Radiomics in breast cancer: Current advances and future directions

Ying-Jia Qi; Guan-Hua Su; Chao You; Xu Zhang; Yi Xiao; Yi-Zhou Jiang; Zhi-Ming Shao

doi:10.1016/j.xcrm.2024.101719

. 2024 Sep 17;5(9):101719. doi: 10.1016/j.xcrm.2024.101719

Radiomics in breast cancer: Current advances and future directions

Ying-Jia Qi ^1,³, Guan-Hua Su ^1,³, Chao You ^2,³, Xu Zhang ², Yi Xiao ^1,^∗, Yi-Zhou Jiang ^1,^∗∗, Zhi-Ming Shao ^1,^∗∗∗

PMCID: PMC11528234 PMID: 39293402

Summary

Breast cancer is a common disease that causes great health concerns to women worldwide. During the diagnosis and treatment of breast cancer, medical imaging plays an essential role, but its interpretation relies on radiologists or clinical doctors. Radiomics can extract high-throughput quantitative imaging features from images of various modalities via traditional machine learning or deep learning methods following a series of standard processes. Hopefully, radiomic models may aid various processes in clinical practice. In this review, we summarize the current utilization of radiomics for predicting clinicopathological indices and clinical outcomes. We also focus on radio-multi-omics studies that bridge the gap between phenotypic and microscopic scale information. Acknowledging the deficiencies that currently hinder the clinical adoption of radiomic models, we discuss the underlying causes of this situation and propose future directions for advancing radiomics in breast cancer research.

Graphical abstract

Qi et al. summarize the workflow of radiomics in breast cancer based on multiple imaging modalities including MG, US, MRI, etc. They also discuss the potential clinical application of radiomics, the prospect of radio-multi-omics, as well as the existing challenges and future directions in integrating radiomics into clinical practice.

Introduction

Breast cancer is one of the most commonly diagnosed cancers and the leading cause of cancer-related mortality in women worldwide, posing a serious threat to their health.¹ The biological heterogeneity of breast cancer, including intertumoral and intratumor heterogeneity, serves as a pivotal contributing factor to the clinical behavior and outcomes.² During the past decades, the clinical outcomes of patients with breast cancer have gradually improved owing to the popularization and standardization of screening and the therapeutic advances.³ Despite this inspiring progress, the precision of the diagnosis, treatment, and prognosis prediction of breast cancer is still unsatisfactory due to the following reasons: (1) insufficient large-scale screening and early detection due to high economic and labor costs and a lack of standardized protocols; (2) inability to rapidly and dynamically monitor the nature and changes of tumor; and (3) unsatisfactory treatment response calling for better prediction models of therapeutic effects. These current situations suggest the significance of further optimizing the clinical workflow of breast cancer utilizing a more powerful accessory model in the era of personalized and precision medicine.

Medical imaging enables the full-scale mapping of intratumoral or peritumoral areas via a non-invasive approach, acting as a mediator bridging the gap between the phenotype and genotype of tumors.⁴ Recently, the conventional diagnosis function of medical imaging may take a step forward with the proposed concept of radiomics.⁵ Radiomics translates medical images into high-throughput mineable data and automatically extracts features to supplement the estimation of clinical indices in various malignancies, including breast cancer.⁶^,⁷ Specifically, while artificial intelligence (AI)-based computer-aided detection/diagnosis (CAD) systems have been widely applied in clinical practice for aiding the screening and diagnosis of breast cancer, radiomics may also have the potential of aiding clinical decision.⁸ In addition, a combination of radiomic and multi-omic approaches, for instance, radiogenomics, may generate a more compelling model for realizing the functions aforementioned, or interpreting the biological significance of the features of medical images.⁹^,¹⁰ Although an increasing number of studies are exploring the potential clinical value of radiomics, most research is still in the initial stages, with few successful clinical applications or United States Food and Drug Administration (FDA) approvals of radiomics reported to date. This suggests that multiple factors restrain the clinical translation of radiomics, awaiting resolution by researchers in the future.¹¹

The current challenges in the clinical practice of breast cancer underscore the importance of utilizing radiomics-based decision support systems and exploring the intricate connections between images and molecular events. In this review, we start from addressing the existing challenges faced in the clinical management of breast cancer. We then detail the imaging modalities, the categories of radiomic features, and the radiomics workflow in breast cancer. In the core section, we summarize the clinical applications of radiomics, including diagnosing, subtyping, staging, and predicting treatment response as well as prognosis in breast cancer. We also decipher the correlation between phenotypical radiomics and microscopic omics data of clinical and scientific significance, namely, “radio-multi-omics.” Finally, we conclude by outlining the current challenges and future prospects for radiomics in the field of breast cancer.

Challenges of the diagnosis and treatment in breast cancer

Epidemiology, diagnosis, and treatment of breast cancer

Breast cancer remains a major global health issue with 2.31 million new cases and over 665,000 deaths in 2022, ranking it high among cancers in both incidence and mortality.¹ Screening is widely regarded as beneficial, reducing incidence and mortality by facilitating early detection and diagnosis.¹² The individualized screening plan should be made according to the stratification of age as well as other risk factors including family history, breast density, specific gene mutations, etc.¹³^,¹⁴ The diagnosis and subtyping of breast cancer typically rely on invasive tissue biopsies, like core needle or fine-needle aspiration, followed by hematoxylin-eosin (H&E) staining and immunohistochemistry (IHC) on surgically removed tissues for histopathological analysis. Clinically, breast cancer classification relies on detecting hormone receptors (HRs) including estrogen receptor (ER) and progesterone receptor (PR), human epidermal growth factor receptor 2 (HER2), and Ki-67 through IHC or in situ hybridization, identifying four clinical subtypes: HR+/HER2−, HR+/HER2+, HR−/HER2+, and triple-negative breast cancer (TNBC).¹⁵ The biological and clinical features vary significantly across subtypes, leading to the clinical application of subtype-specific therapeutic methods aimed at improving prognosis.¹⁶ In particular, endocrine therapy is especially effective for treating HR+/HER2− breast cancer, while therapies targeting HER2 benefit patients with HER2+ breast cancer. TNBC, characterized by the absence of HR and HER2 expression on tumor cells, is the most invasive and heterogeneous subtype. While chemotherapy remains the standard treatment, emerging therapies such as immunotherapy, antibody-drug conjugates, and poly ADP-ribose polymerase inhibitors (PARPi) are gaining prominence.¹⁷

Current challenges in breast cancer

The current challenges in the diagnosis and treatment of breast cancer lie in the insufficient large-scale screening, the inability to rapidly and dynamically monitor the nature and changes of tumors, and the lack of precise treatment options. Breast cancer screening remains inadequate due to high economic and labor costs, as well as the absence of standardized protocols. The traditional dependence on radiologists for medical image interpretation not only contributes to these costs but also affects the accuracy and reproducibility of results due to variable expertise and heavy workloads. Additionally, inconsistencies in risk assessment criteria and age-specific recommendations across various screening guidelines cause confusion among both potential patients and clinical doctors.¹²^,¹⁸^,¹⁹ These issues undermine the primary objective of achieving early detection and diagnosis through large-scale screening.

Moreover, current methods, primarily tissue biopsy including core needle biopsy and fine-needle aspiration, for revealing the nature of lesion or continuously and dynamically monitoring tumor changes during neoadjuvant therapy (NAT) are still inadequate. Tissue biopsy, typically conducted once at baseline, is unsuitable for monitoring tumor changes during NAT. Furthermore, these techniques fail to capture pathological changes across the entire tumor, increasing the risk of missing lesions in unsampled areas. Consequently, there is a pressing need for comprehensive methods to consistently assess the nature and changes of breast cancer.

Although subtyping-based treatment has improved the prognosis for many patients with breast cancer, some still do not benefit from targeted treatment options. Various multigene assays for prognosis prediction or treatment options in breast cancer are developed, including 12-gene EndoPredict for predicting distant recurrence in patients with HR+/HER2− breast cancer treated with adjuvant endocrine therapy, 21-gene Oncotype DX for chemotherapy decision-making in patients with ER+, node-negative breast cancer, PAM50-based Prosigna for subtyping and recurrence prediction, and 70-gene MammaPrint for prognosis prediction in early-stage breast cancer.²⁰^,²¹^,²²^,²³ In addition, various biomarkers for predicting treatment response in patients have been identified. For instance, tumor-infiltrating lymphocytes (TILs) identified by H&E slices or digital pathology, as well as the expression of certain molecules, including programmed cell death ligand-1 (PD-L1), have been utilized for predicting the response to NAT in patients with TNBC.²⁴ However, these assays or biomarkers only predict the efficacy of treatment in a moderate proportion of patients, suggesting the insufficiency of the current prediction methods. Therefore, to achieve precise treatment of breast cancer, there is an urgent need for more effective indicators or combinations thereof to develop satisfactory predictive models.

Radiomics in breast cancer

In the context of breast cancer, radiomics refers to extracting innumerable quantitative features from specific medical imaging modalities using high-throughput computing, followed by the mining of the relevance between these features and the clinicopathological indices or multi-omics analysis of breast cancer.²⁵^,²⁶ In this section, we offer a concise overview of prevalent imaging modalities and categories of radiomic features, along with an illustration of the workflow of machine learning (ML)-based as well as deep learning (DL)-based radiomics in breast cancer.

Imaging modalities

The essential imaging modalities utilized in the clinical practice for breast cancer include mammography (MG), ultrasound (US), magnetic resonance imaging (MRI), and positron emission tomography/computed tomography (PET/CT), which are also the most common modalities for radiomic studies. Each modality has unique characteristics and best-matched application conditions. MG is most commonly applied in screening worldwide, partly because of its convenience and inexpensiveness, and it is especially sensitive for detecting microcalcifications. Digital breast tomosynthesis, also known as 3D MG, improves upon the two-dimensional image provided by traditional MG and enhances cancer detection rate.²⁷ While MG exhibits relatively low sensitivity in dense mammary glands, contrast-enhanced mammography (CEM) effectively addresses this issue.²⁸ US serves as a valuable adjunct to MG in patients with dense breast tissue, possessing the capability to differentiate between cystic and solid lesions.²⁹ Shear wave elastography (SWE) provides richer and more definitive information than conventional US and color Doppler imaging, by observing the stiffness of lesions.³⁰ Recently developed automated breast US tends to overcome the limitation of operator dependency of traditional handheld US and also provides 3D reformatting images for higher diagnostic accuracy.³¹ MRI is the most sensitive imaging modality among all and is particularly accurate in detecting invasive diseases, while dynamic contrast-enhanced MRI (DCE-MRI) is the basis for clinical applications.³² The MRI parameters used to generate radiomic features, such as T1-weighted imaging (T1-WI) conventionally applied for revealing the anatomical structures, T2-weighted imaging (T2-WI) clearly showing edema and fluid therefore better indicating significant lesions, diffusion-weighted images (DWIs) for uncovering the diffusion status of water molecules, apparent diffusion coefficient mapping based on DWI for describing the velocity and range of diffusion in different directions, DCE-MRI that obtains images before and after the injection of the contrast agent through continuous, repetitive, and rapid scanning to evaluate the structure and function of the microvessels, etc., each provide unique radiology information. ¹⁸F-Fluorodeoxyglucose PET/CT combines the glucose metabolic and morphological features of tumors, reflecting the biology of malignant cells at the molecular level.³³ The conventional positron emission tomography parameters include the standardized uptake value (SUV)_max, SUV_mean, SUV_peak, metabolic tumor volume, and total lesion glycolysis (TLG).

In the development of radiomic models, while some studies have focused on extracting features from single modalities, primarily using MG, US, or DCE-MRI, there is a growing trend toward building multimodal radiomics models. These models aim to harness the unique strengths of various imaging techniques and enhance overall efficacy. They may incorporate multiple sequences from a single technique, such as multiparametric MRI (mpMRI), which includes T1-WI, T2-WI, DWI, and DCE-MRI for instance, or they may combine different modalities, such as PET/CT and US.³⁴^,³⁵

Categories of radiomic features

According to AI methodologies, quantitative radiomic features extracted from images can be divided into two types, hand-crafted features generated by ML and DL features.

Hand-crafted features can be further divided into semantic features and agnostic features.⁶ Semantic features, also referred to as morphological features, refer to the explicable features possibly used by radiologists to describe tumor features, including size, shape, and tumor surface features such as surface area, surface-to-volume ratio, sphericity, compactness, and convexity.³² Agnostic features are often beyond comprehension, including first-order, second-order, and higher-order statistical features.⁶ First-order statistical features, namely, histogram features, are the simplest statistical features describing the distribution of voxels based on the global gray-level histogram regardless of spatial distribution, such as gradient mean, variance, randomness (entropy), kurtosis (asymmetry), and skewness (flatness).³⁶ Second-order statistical features, namely, textural features, are able to measure intratumor heterogeneity (ITH) by describing the correlation of spatially adjacent voxels, such as the gray-level co-occurrence matrix (GLCM), the gray-level run length matrix, the gray-level size zone matrix, and the neighborhood gray-level different matrix.³⁷^,³⁸ Higher-order statistical features impose filter grids on images for repetitive or nonrepetitive patterns, including Minkowski functionals, wavelets, and Laplacian transforms of Gaussian band-pass filters, etc.⁶

As aforementioned, CAD systems, primarily based on DL, have improved efficacy and are increasingly being adopted for clinical applications in breast cancer screening.³⁹ There is an accepted great advantage of DL, which is based on multi-layered artificial neural networks to solve highly complex problems, capable of automatically identifying and learning complex patterns in imaging data.⁴⁰ Recent advances in DL have demonstrated that neural networks can automatically extract radiomics features without the need for human intervention, leading to improved prediction performance.⁴¹ This emerging field is referred to as DL radiomics. Despite being a novel field, DL radiomics capitalizes on the learning capabilities of DL and the interpretability of radiomics, presenting significant potential for development. The extraction of DL features does not rely on traditional hand-crafted methods, instead deriving directly from raw images through deep neural networks. The types of DL features include, but are not limited to, outputs extracted from various convolutional neural network (CNN) layers, including convolutional layers, average pooling layers, and fully connected layers.⁴²^,⁴³^,⁴⁴ Compared to hand-crafted features, these DL features possess a more abstract meaning; therefore, they are usually not categorized as hand-crafted features.

Workflow of radiomics in breast cancer

Both ML and DL radiomics comprise four major steps: image acquisition, tumor segmentation, feature extraction, and model building and analysis (Figure 1). In the radiomics workflow, the primary distinction between ML and DL radiomics lies in the feature extraction process. For assessing the quality of radiomics studies, Lambin et al. initiated the Radiomics Quality Score in 2017 by assigning points according to 16 different items for a maximum score of 36.⁴⁵ This evaluation system has been widely accepted and employed in various studies, mainly in systematic reviews.⁴⁶^,⁴⁷

Image acquisition

Collecting medical images with quantitative data from multiple imaging modalities including MG, US, MRI, and PET/CT, is the first step of radiomics. These initial two-dimensional images can be reconstructed into three dimensions in preparation for subsequent feature extraction.³² Standardized imaging protocols and parameters, including scanner equipment, acquisition techniques and software, reconstruction parameters, and contrast administration, are essential for eliminating the discrepancy of variables of different medical centers.⁴⁸ Standardization and calibration are more difficult for non-ionizing techniques (US and MRI) than photon detection techniques (MG and PET/CT). Currently, the utilization of standardized protocol in radiomic studies is still inadequate.⁶ However, efforts are being made to improve standardization. The Quantitative Imaging Network (QIN) initiated by the National Cancer Institute suggested the improvement of data integration methods from diverse platforms for standardized quantitative imaging protocols and technical standards, with the goal of developing and implementing robust clinical decision support tools.⁴⁹ In addition, the QIN suggests delving into the published public research resources as well as data sharing, and The Cancer Imaging Archive database was selected as the official repository for sharing QIN data.⁵⁰ In addition, efforts to establish standards for biomarker development, validation, and implementation were made by the Quantitative Imaging Biomarkers Alliance, which was launched by the Radiological Society of North America and provides scan protocols and software for the standard analysis of quantitative images.⁵¹

Tumor segmentation

Next, with the purpose of identifying tumors for further information extraction, the region of interest (ROI) is delineated artificially or (semi)automatically. The segmentation of two-dimensional images, such as MG and US images, is manipulated by sketching it on a single graph. For imaging modalities such as computed tomography (CT), MRI, and PET/CT, a combination of two-dimensional ROI layers facilitates showing ROIs in a three-dimensional form, also named volume of interest. Segmentation methods should be reproducible and reliable; however, precisely defining the tumor area is challenging.²⁵ Manual segmentation is time-consuming, and the definition of the tumor margin varies greatly among different radiologists, introducing unexpected bias.²⁵ Semiautomatic segmentation is more satisfactory, with the workflow of automated segmentation based on various algorithms with experienced radiologists aiding the correction.⁵² Free and publicly available platforms such as 3D-Slicer (www.slicer.org) and ITK-SNAP (www.itksnap.org) can be employed for semiautomatic segmentation for quantitative image feature extraction.⁵³ Several DL-based automatic segmentation methods developed in various studies have proven robust; however, most remain under research and have not yet gained the widespread acceptance and usage enjoyed by manual or semiautomatic segmentation methods.⁵⁴^,⁵⁵ Besides whole tumor segmentation as ROIs, some studies have utilized habitat imaging to further segment subregions within the tumor, extending the tumor boundaries to segment the peritumoral areas in images, thereby providing additional dimensions for radiomic feature extraction.⁵⁶

Feature extraction

After segmentation, radiomic features, as summarized earlier, are extracted globally or locally from ROIs.⁵² Before that, pre-processing and reconstruction of the images is necessary.⁴⁵ The goal of pre-processing is to eliminate noise by signal filtering, intensity discretization, or resampling within ROIs. Representative filters include the Laplacian of Gaussian and wavelet transform.⁵⁷^,⁵⁸ Categories of ML and DL features are summarized earlier. Owing to the distinction between ML and DL radiomics in the feature extraction process, we will discuss them separately.

For ML radiomics, Conti et al. summarized abundant user-friendly and open-access tools for feature extraction, among which the open-source python package Pyradiomics is the most commonly used tool.²⁶ For feature standardization, Zwanenburg et al. initiated a three-phase assessment guideline for imaging biomarker standardization, named the image biomarker standardization initiative.⁵⁹ It is worth noting that there is a difference between feature extraction from two-dimensional ROIs and three-dimensional ROIs; for instance, the texture analysis of three-dimensional ROIs is based on extended, multisort co-occurrence matrices, which may contain more information than two-dimensional textures.⁶⁰

While DL models operate as a “black box” without a separate step of feature extraction, DL radiomics enhances interpretability by a radiomic-specific step of extracting DL features from the outputs of CNN layers. Different with ML process, which directly extracts features from original images, an end-to-end DL model has already been trained and fine-tuned to reach satisfactory performance before feature extraction, which allows the best and most related features to be extracted. Similar to that of ML, the three-dimensional CNN, compared to traditional two-dimensional image analysis, considers the three-dimensional context, thereby capturing more comprehensive information across various levels of a solid tumor and enabling more accurate decision-making.⁶¹

Model building and analysis

The last part of radiomic workflow is composed of three major procedures: feature selection, modeling, and model validation. First, tens of thousands of features can be extracted from images; however, only some of them are relevant to the prediction model, while others may be irrelevant or cause overfitting. Therefore, features should be selected for building a solid statistical model. Methods of feature selection include filter, wrapper, embedding, and their hybridization.²⁵ Filter and wrapper methods generate a subset of variables before modeling starts. However, for variable measurement, filter methods evaluate them independent of the model, while wrapper methods measure the significance of variables according to the performance of the model.⁶² Filter methods, as feature-ranking methods using a scoring criterion, can be divided into two categories: univariate methods and multivariate methods. The score criterion of the former only depends on the feature relevancy ignoring the feature redundancy, while the latter is a weighted sum of feature relevancy and redundancy.⁶³ In embedding methods, both the generation and evaluation of variables are performed during the construction of the model.

Modeling is the core process among all the steps in radiomic analysis. Strategies for modeling include supervised, semi-supervised, and unsupervised learning methods. Supervised learning requires clinical labels to learn the relationships between labels and features, and specific machine learning algorithms include linear discriminant analysis (LDA), support vector machine (SVM), least absolute shrinkage and selection operator (LASSO), random forest (RF), logistic regression (LR), k-nearest neighbors, etc.⁶⁴^,⁶⁵^,⁶⁶ Unsupervised learning methods are not dependent on clinical labels and are performed by clustering based on distance measurements, including k-means clustering, fuzzy clustering, consensus clustering, denoising autoencoder (DA), generative adversarial networks (GANs), etc.⁶⁷^,⁶⁸^,⁶⁹^,⁷⁰^,⁷¹ Supervised learning has stronger performance but requires a large amount of data, while unsupervised learning provides a possibility for clustering without labeled data, but the performance may be unsatisfactory. Semi-supervised learning, as an intermediate method, uses a large amount of unlabeled data for information mining and small amounts of labeled data to determine the relationships between features and clinical labels; representative models include transductive SVM and semi-supervised GANs.⁷²^,⁷³ Models based on two-dimensional imaging modalities are rather cost-effective but may not fully capture complex pathological changes, while models based on three-dimensional modalities with more complex procedures typically provide higher reliability and more comprehensive disease assessments. The established models may serve as a classification or clinical outcome predictor, as well as a molecular or biological alteration predictor. Detailed applications are introduced in the following sections.

Validation in an internal dataset is necessary for all models, and validation (testing) in independent external datasets is preferable for higher credibility by detecting and precluding overfitting.⁷⁴ As the most frequently used internal validation method, k-fold cross validation, which splits datasets into different subsets for training and validating, can decrease the chance of overfitting before external validation.⁷⁵ The indices for measuring the performance of classification models include the area under the curve (AUC), sensitivity, specificity, and accuracy. The Harrell’s concordance index (C-index) and time-dependent receiver operating characteristic curves are commonly employed to assess the predictive ability of prognosis-related variables for survival.⁷⁶^,⁷⁷

Utilizing radiomics to predict breast cancer clinicopathological indices and clinical outcomes

The potential application of radiomics in the clinical practice of breast cancer mainly consists of two aspects: (1) classification of tumor, i.e., discriminating benign and malignant lesions, molecular subtypes, and other clinicopathological indices including sentinel lymph node status and (2) prediction of treatment response to NAT and clinical outcomes including survival and recurrence. Research exploring the clinical applications of radiomics in breast cancer is systematically summarized in Table 1.

Table 1.

Summary of radiomic studies predicting breast cancer clinical indices

Task	Number of patients	Image modalities	Number and type of selected features	Classifiers	Best performance	Reference
Classification: malignant vs. benign	N = 182 (LOOCV)	FFDM	Tumor: 4, including spiculation, margin sharpness, size, and circularity; Parenchymal: 4, including edge gradient, contrast, power law beta, and 2 GLCM	BANN	Tumor + parenchymal features: AUC = 0.84; Tumor features: AUC = 0.79; Parenchymal features: AUC = 0.67	Li et al.⁷⁸
	Training: N = 850; Testing: N = 212 (internal) + N = 279 (external)	CEM	Hand-crafted features; Clinical features: 8, including number of pregnancies, age, family history, etc.	XGBoost	Automatic segmentations, hand-crafted features + DL model: AUC = 0.88, sen = 0.90, spe = 0.86	Beuque et al.⁵⁵
	MG: described in a previous study; 3CB: training: N = 45, testing: N = 109	MG, 3CB	MG: 5, including lesion size, average gray value, etc.; 3CB: 4, including the median water thickness within a mass, etc.	LDA	MG + 3CB features: sen = 0.97, spe = 0.51, PPV = 0.49, NPV = 0.97, 35.8% fewer total biopsies	Drukker et al.⁷⁹
	Training: N = 95; Testing: N = 127 (external)	mpMRI including AKC and ADC	Hand-crafted features: 19, including 3 ADC and 16 AKC; Clinical feature: 1, age	RF	All breast cancer: AUC = 0.91, sen = 0.98, spe = 0.70; BI-RADS 4a subgroup: AUC = 0.95, sen = 1.00, spe = 0.74;	Bickelhaupt et al.⁸⁰
	Training: N = 159; Testing: N = 63 (internal) + N = 50 (external)	mpMRI	Hand-crafted features: 3, including ADC median, T2-WI range, and SER GLCM joint entropy; Clinical feature: 1, BI-RADS	SVM, LR	Radiomics score: AUC = 0.95, sen = 0.90, spe = 0.92, acc = 0.91; Nomogram: AUC = 0.98, sen = 0.92, spe = 0.92, acc = 0.92	Zhang et al.⁸¹
	Training: N = 447; Validation: N = 578	mpMRI	Hand-crafted features with L1 regularization: 47, including 6 from T2-WI, 1 shape feature (sphericity), 39 from dynamic sequences	L1 regularization, PCA	ML with L1 regularization: AUC = 0.81, sen = 0.73, spe = 0.76; DL: AUC = 0.88, sen = 0.78, spe = 0.85; Human: AUC = 0.98, sen = 1.00, spe = 0.86	Truhn et al.⁸²
Classification: molecular subtypes	N = 91	T1-weighted DCE-MRI	Hand-crafted features: 38, including 10 first order, 10 enhancement texture, 14 kinetics, and 4 enhancement kinetics	N.A.	ER + vs. ER−: AUC = 0.89; PR + vs. PR−: AUC = 0.69; HER2+ vs. HER2−: AUC = 0.65; TNBC vs. others: AUC = 0.67	Li et al.⁸³
	Training: N = 91; Testing: N = 52 (external)	mpMRI	N.A. for selected features	LDA	Luminal B vs. luminal A: acc = 0.84; Luminal B vs. TNBC: acc = 0.84; Luminal B vs. all others: acc = 0.89; HER2-enriched vs. all others: acc = 0.81	Leithner et al.⁸⁴
	N = 180	CEM	18 hand-crafted features for HER2; 28 hand-crafted features for HR; 24 hand-crafted features for luminal vs. non-luminal	LR, LASSO	HER2+ vs. HER2−: AUC = 0.77; ER+ vs. ER−: AUC = 0.80; luminal vs. non-luminal: acc = 0.94	Petrillo et al.⁸⁵
	N = 1,506	T1-weighted DCE-MRI	Hand-crafted features for tumors classified as luminal A by IHC and luminal B by St Gallen: 6, including sphericity, irregularity, etc.	Mann-Whitney U test	IHC as reference standard: AUC = 0.66; St Gallen as reference standard: AUC = 0.62; IHC and St Gallen agreement as reference standard: AUC = 0.74	Ji et al.⁸⁶
Classification: molecular subtypes and clinicopathological indices	N = 124 (5-fold CV)	18F-FDG PET/MRI	6 hand-crafted features each model, including first order, GLCM, NGTDM, NGLDM, GLSZM, and GLRLM features	LASSO	Luminal B vs. luminal A: AUC = 0.98, acc = 0.97; Luminal vs. others: AUC = 0.95, acc = 0.89; Ki-67: AUC = 1.00, acc = 0.96; Grading: AUC = 0.78	Umutlu et al.⁸⁷
Classification: HER2 status	Training: N = 177; Testing: N = 162	MDCT	Hand-crafted features: 7, including 3 GLCM, 3 GLSZM, and 1 GLRLM	SVM, multivariate LR	HER2, hand-crafted features: C-index = 0.70; HER2, DL model: C-index = 0.78; HER2, combined: C-index = 0.81	Yang et al.⁴¹
Classification: TNBC vs. non-TNBC	Training: N = 430; Testing: N = 430 (internal) + N = 164 (external)	DCE-MRI	TNBC: 11 features including 5 peritumoral, 3 intratumoral, 1 entropy, and 2 tumor-peritumoral features	LR, LASSO, SVM	TNBC vs. others: AUCs = 0.82, 0.72, and 0.61 in internal and two external datasets	Jiang et al.⁸⁸
Classification: ALN metastasis positive vs. negative	Training: N = 849; Testing: N = 365	mpMRI	Hand-crafted features: 30 each model, including first order and texture features; Clinical features, including age, T and N stage, histological grade, HER2 status, etc.	LASSO-LR, LASSO-SVM, RF-LR, RF-SVM	ALN metastasis, nomogram: AUC = 0.90; DFS, nomogram: AUCs = 0.87, 0.90, and 0.89 for 1-year, 2-year, and 3-year DFS	Yu et al.⁸⁹
	Training: N = 279; Testing: N = 132	T1-weighted DCE-MRI	Hand-crafted features: 12, including 3 first-order, 1 shape, 4 GLCM, 3 GLSZM, and 1 GLRLM; Clinical features: 2, including MRI-reported LN status and LN palpability	LASSO, LR, SVM	ALN metastasis, radiomic features: AUC = 0.78, sen = 0.78, spe = 0.72, acc = 0.70; ALN metastasis, radiomic features: AUC = 0.87, sen = 0.85, spe = 0.80, acc = 0.83	Han et al.⁹⁰
	N = 166 (5-fold CV)	mpMRI	Model 1: 4 Ipris features; Model 2: 6 intratumoral features; Model 3: 2 Ipris features intratumoral features	Spearman correlation analysis, SVM-RFE	Model 1: AUC = 0.83, sen = 0.89, spe = 0.70, acc = 0.74; Model 2: AUC = 0.82, sen = 0.87, spe = 0.73, acc = 0.80; Model 3: AUC = 0.86, sen = 0.91, spe = 0.77, acc = 0.69	Zhan et al.⁹¹
	Training: N = 237; Testing: N = 106	US	Hand-crafted features: 12, including 2 shape, 1 first order, 1 GLCM, 3 GLSZM, 3 NGTDM, and 2 GLDM; Clinical features: 2, including age and largest lesion diameter	LASSO	Radiomic features: AUC = 0.83; Nomogram: AUC = 0.72, sen = 0.70, spe = 0.75, acc = 0.73	Gao et al.⁹²
	Training: N = 234; Testing: N = 723 (external) + N = 81 (prospective)	mpMRI	Stacking model: 33 hand-crafted features, including 9 pre-NAC, 10 post-NC, and 11 delta-NAC	LASSO, SVM, LR, RF	External: acc = 0.90; Prospective: acc = 0.97	Zhu et al.⁹³
	Training: N = 466; Testing: N = 118	US + SWE	DL features; Clinical features	ResNet50, SVM	N0 vs. others, nomogram: AUC = 0.90, sen = 0.82, spe = 0.84, acc = 0.81; N1-2 vs. N ≥ 3, nomogram: AUC = 0.91, sen = 0.85, spe = 0.87, acc = 0.80	Zheng et al.⁴³
	Training: N = 621; Testing: N = 262 (external)	US	Hand-crafted features: 4; DL features: 45; Clinical features: 4	ResNet50, LASSO	Nomogram: AUCs = 0.914, 0.929, and 0.952 in the three external validation cohorts	Liu et al.⁴⁴
	Training: N = 1,737; Testing: N = 1,397 (retrospective) + N = 478 (prospective)	CT	Hand-crafted features: 36	K-means	Retrospective: average AUC = 0.85; Prospective: average acc = 0.87	Qu et al.⁹⁴
Prediction: NAC response pCR vs. non-pCR	Training: N = 128; Testing: N = 286 (external)	mpMRI	Hand-crafted features: 26, including 7 from T2-WI, 8 from ADC maps, 3 from T1+C, and 8 from mpMRI; Clinical features: 2, including PR and HER2 status	SVM	Hand-crafted features from mpMRI: AUCs = 0.79, 0.87, 0.79, and 0.84 in all breast cancer, HR+ HER2−, HER2+, and TNBC subgroups; Nomogram: AUC = 0.80	Liu et al.⁹⁵
	Training: N = 78; Testing: N = 39	T1-weighted DCE-MRI	Approximately 10 features each model	LDA, DLDA, naive Bayes, SVM	DLDA: AUC = 0.78; DLDA, HR+ HER2− subgroup: AUC = 0.78; Naive Bayes, TNBC or HER2+ subgroup: AUC = 0.78	Braman et al.⁹⁶
	Training: N = 171; Testing: N = 78	DCE-MRI	5 hand-crafted features each region subgroup, first order	Bhattacharyya distance, DLDA	Intratumoral + peritumoral features: AUC = 0.80, sen = 0.94, spe = 0.58, acc = 0.79	Braman et al.⁹⁷
	N = 79 (10-fold CV)	18F-FDG PET/CT	4 models including 1–2 hand-crafted features, 2 PET-derived features (SUV_max and TLG), and 3 clinical features (age, molecular subtype, and HER2 status)	LR, LASSO	AUC = 0.70 to 0.73 in four models	Antunovic et al.⁹⁸
	Training: N = 310; Testing: N = 134 (internal) + N = 81 (external)	US	Hand-crafted features: 9, including 1 shape, 1 first order, 3 GLCM, 3 GLSZM, and 1 GLDM; Clinical features: 5, including N stage, ER status, Ki-67, and long and short axis of lesion	MRMR, LASSO	Radiomic + clinical features: AUC = 0.86, sen = 0.87, spe = 0.48, acc = 0.73; Radiomic features: AUC = 0.72, sen = 0.73, spe = 0.86, acc = 0.78	Zhang et al.⁹⁹
	Training: N = 81; Testing: N = 37	CESM	Hand-crafted features: 32, including 2 first order, 10 GLCM, 6 GLSZM, 7 GLRLM, 1 NGTDM, and 5 GLDM	Mann-Whitney U test, LASSO	Intratumoral regions + 5-mm peritumoral regions: AUC = 0.85	Mao et al.¹⁰⁰
	Training: N = 356; Testing: N = 236 (external)	US	Hand-crafted features: 12, including 5 from pre-treatment images and 7 from post-treatment images; DL features: 64 Clinical features: 2, including N stage and PR status	LR, LASSO, DenseNet-201	DL + hand-crafted features: AUC = 0.94, sen = 0.89, spe = 0.81, acc = 0.84; DL + hand-crafted features: AUCs = 0.90, 0.95, and 0.93 for HR+ HER2−, HER2+, and TNBC subgroups	Jiang et al.¹⁰¹
	Training: N = 409; Testing: N = 853 (external)	mpMRI	15, 20, and 13 features for HR+ HER2−, HER2+, and TNBC subgroups, including textural and DL features	ResNet50, LR, RF, XGBoost, SVM, MLP	MLP, DL + hand-crafted features: AUCs = 0.96, 0.97, and 0.96 for HR+ HER2−, HER2+, and TNBC subgroups	Huang et al.¹⁰²
	Training: N = 286; Testing: N = 281 (external)	US	Pre-NAC: 3 hand-crafted and 3 DL features; Post-NAC: 3 hand-crafted and 5 DL features	LASSO, RF, SVM	Pre-NAC model: AUC = 0.79; Post-NAC model: AUC = 0.85; Fusion model: AUC = 0.90	Fu et al.¹⁰³
Prediction: NAC response RCB 0, I, II, and III	Training: N = 335; Testing: N = 713 (external)	mpMRI	Model 1: 35 hand-crafted and DL, pre-NAC, mid-NAC, and delta-NAC features Model 2: 40 hand-crafted and DL, pre-NAC, mid-NAC, and delta-NAC features	ResNet50, LASSO, LR, SVM, RF	Model 1 (RCB 0–II vs. III): AUCs = 0.91 to 0.94, acc = 0.92 to 0.94; Model 2 (RCB 0 and I vs. II and III): AUCs = 0.90 to 0.92, acc = 0.82 to 0.84;	Li et al.¹⁰⁴
Prediction: DFS of patients with invasive breast cancer	Training: N = 194; Testing: N = 100	mpMRI	Hand-crafted features: 156, including shape and GLCM features Clinical features: including N stage, molecular subtype, etc.	Elastic net Cox proportional regression	Nomogram: C-index = 0.76; Radiomic features: C-index = 0.64; Clinical features: C-index = 0.72	Park et al.¹⁰⁵
Prediction: DFS of patients with locally advanced breast cancer	Training: N = 139; Testing: N = 139 (internal) + N = 91(external)	T1-weighted DCE-MRI	Hand-crafted features: 15, list not provided; Clinical features: including age, cTNM stage, etc.	LASSO	Nomogram: C-index = 0.61; Radiomic features: C-index = 0.61	Wang et al.¹⁰⁶
Prediction: OS and RFS of TNBC	N = 202	T1-weighted DCE-MRI	Related to prognosis: 1, peritumoral variance of dependence non-uniformity (Peri_V_DN)	LR, SVM, LASSO	Peri_V_DN, OS: p = 0.004; Peri_V_DN, RFS: p = 0.01	Jiang et al.⁸⁸
Prediction: DFS of patients with early-stage breast cancer	Training: N = 849; Testing: N = 365	mpMRI	Hand-crafted features: 30 each model, including first order and texture features; Clinical features, including tumor number, T and N stage, histological grade, PR status, etc.	LASSO-Cox regression, RF-Cox regression	Hand-crafted features, RF-Cox regression: AUCs = 0.80, 0.83, and 0.81 for 1-year, 2-year, and 3-year DFS; Nomogram: AUCs = 0.89, 0.91, and 0.90 for 1-year, 2-year, and 3-year DFS	Yu et al.⁸⁹
Prediction: recurrence risk of patients with breast cancer	Training: N = 127; Testing: N = 60	T1-weighted DCE-MRI	Hand-crafted features: 4; BPE features: 1 contralateral and 3 ipsilateral	LASSO	Nomogram: AUC = 0.79	Arefan et al.¹⁰⁷

Open in a new tab

MG, mammography; FFDM, full-field digital mammography; DBT, digital breast tomosynthesis; CEM, contrast-enhanced mammography; CESM, contrast-enhanced spectral mammography; 3CB, three-compartment breast; mpMRI, multiparametric magnetic resonance imaging; DCE-MRI, dynamic contrast-enhanced magnetic resonance imaging; AKC, apparent kurtosis coefficient; ADC, apparent diffusion coefficient; US, ultrasound; SWE, shear wave elastography; PET, positron emission tomography; CT, computed tomography; MDCT, multidetector computed tomography; BPE, background parenchymal enhancement; CV, cross validation; LOOCV, leave-one-out cross-validation; GLCM, gray-level co-occurrence matrix; NGLDM, neighboring gray-level dependence matrix; NGTDM, neighboring gray-tone difference matrix; GLRLM, gray-level run length matrix; GLSZM, gray-level size zone matrix; GLZLM, gray-level zone length matrix; GLDM, gray-level dependence matrix; ALN, axillary lymph node; SLN, sentinel lymph node; IHC, immunohistochemistry; LVI, lymphovascular invasion; Ipris, intra-peritumoral textural transition; SUV, standardized uptake value; TLG, total lesion glycolysis; ER, estrogen receptor; PR, progesterone receptor; HER2, human epidermal growth factor receptor 2, TNBC, triple-negative breast cancer; DL, deep learning; BANN, Bayesian artificial neural network; LDA, linear discriminant analysis; DLDA, diagonal linear discriminant analysis; RF, random forest; SVM, support vector machine; LR, logistic regression; PCA, principal component analysis; RFE, recursive feature elimination; MRMR, minimal redundancy maximum relevance; LASSO, least absolute shrinkage and selection operator; MLP, multi-layer perception; AUC, the area under the curve; sen, sensitivity; spe, specificity; acc, accuracy; PPV, positive predictive value; NPV, negative predictive value; NAC, neoadjuvant chemotherapy; pCR, pathological complete response; RFS, recurrence-free survival; OS, overall survival; DFS, disease-free survival; N.A., not available.

Distinguishing benign masses versus malignant tumors

As mentioned earlier, DL-based CAD systems are becoming an increasingly favorable trend in real-world clinical practice, enhancing detection rates and diminishing unnecessary recalls, as summarized and discussed thoroughly by Geras et al. and Diaz et al.¹⁰⁸^,¹⁰⁹ Although there hasn’t been clinical applications yet, there are also studies exploring the possibility of realizing radiomics-based screening across large populations to determine the biological nature of lesions displayed in medical images. MG is the most common screening technology for discriminating the nature of lesions; therefore, most radiomic models in this field are MG based. The most meaningful regions of the breast for feature extraction are identified in various studies. For instance, Mudigonda et al. reported that malignant tumor exhibits apparently diverse gradient and GLCM features especially at the tumor boundary compared to benign nodules.¹¹⁰ Li et al. found that combining features of the normal parenchyma from the contralateral breast with lesions may improve the accuracy of lesion classification (AUC = 0.84 vs. 0.79).⁷⁸ Advanced MG technologies, such as CEM, have also been further studied for building radiomic models for classification.⁵⁵ Interestingly, Drukker et al. established a model by extracting features from automatically segmented MG and quantitative three-compartment breast images in a prospective clinical dataset, followed by 10-fold cross-validations with LDA. The positive predictive value of the combined radiomics model outperformed conventional diagnostic digital MG (0.49 vs. 0.32). Besides, the combined model may reduce total biopsies by 35.8% compared to conventional MG.⁷⁹ In addition to MG, MRI-based radiomic models also assist in screening, especially in cases of suspicious lesions with BI-RADS (Breast Imaging-Reporting and Data System) 4a or 4b according to MG.⁸⁰ Zhang et al. built a model based on mpMRI that reached an AUC of 0.95, while combining the radiomics score and BI-RADS score further enhanced the AUC to 0.98.⁸¹ However, the exploration of DL radiomics remains underexplored in this field. In conclusion, various ML-based radiomic models, primarily utilizing features extracted from MG and MRI, have proven effective in distinguishing malignant tumors from benign lesions, but the potential of DL radiomics warrants further exploration.

Classification of molecular subtypes and other clinicopathological indices

Radiomics is also widely studied for the classification of molecular subtypes based on ER, PR, HER2, and Ki-67 expression, as well as other clinicopathological indices, including histological grade and lymphovascular invasion (LVI) of breast cancer. Li et al. built one of the first classification models based on MRI. The models of ER+ vs. ER−, PR + vs. PR−, HER2+ vs. HER2−, and triple-negative vs. other subtypes achieved AUCs of 0.89, 0.69, 0.65, and 0.67, respectively. The study also identified enhancement texture (entropy) as a feature highly correlated with subtyping.⁸³ The performance of MRI-based radiomic models has improved over time. For instance, the accuracies of a more recent classification model based on DCE-MRI achieved over 80% in each comparison group, while the best differentiation groups were luminal B, followed by HER2-enriched cancers.⁸⁴ In addition, Mazurowski et al. identified the characteristics of luminal B cancers, that is, a higher ratio of lesion enhancement rate to background parenchymal enhancement rate.¹¹¹ For other imaging modalities, a CEM-based prediction model for discriminating luminal versus non-luminal tumor with 15 features extracted by cranium caudal view reached an accuracy of 0.94.⁸⁵ Besides intratumoral regions, features extracted from peritumoral regions may also contain information for classification; for instance, Jiang et al. combined 11 features, including 5 extracted from peritumoral regions, to construct a model for TNBC classification, reaching the highest AUC of 0.72 in the external datasets.⁸⁸ In conclusion, models based mainly on radiomic features extracted from MRI, as well as from other modalities, are capable of predicting the status of single vital biomarkers, including ER, PR, HER2, and Ki-67, for the prediction of the molecular subtypes, and other clinicopathological indices including histological grade and LVI of breast cancer.

Prediction of axillary and sentinel lymph node status

Research in radiomics also aims to predict sentinel lymph node (SLN) status, thereby reducing the need for invasive biopsies and reducing unnecessary complete axillary lymph node (ALN) dissections. Various radiomic studies have shown that nomograms combining radiomic and clinical features can satisfactorily predict ALN or SLN status, whereas the majority of radiomic studies have focused on MRI as the imaging modality. Typically, in a multicenter study including 1,214 patients, a clinical-radiomic nomogram based on DCE-MRI for the preoperative identification of ALN metastasis status was established with an LASSO-LR classifier, which received the highest AUC of 0.90.⁸⁹ To demonstrate the better performance of nomograms compared with radiomics-only models, Han et al. developed two models with features only from T1-DCE images and a nomogram based on radiomic and clinical features for distinguishing the number of metastatic LNs, reaching AUCs of 0.78 and 0.87, respectively.⁹⁰ In addition, a recent study created a 3D atlas based on enhanced lung-enhanced CT for building a prediction model for ALN status, followed by a prospective testing with a dataset of 128 negative and 350 positive patients, reaching AUCs of 0.86 and 0.87.⁹⁴

In terms of DL radiomics, models with US-based features are the main focus. For instance, Zheng et al. combined DL features and clinical parameters based on conventional US and SWE, achieving AUC of 0.902 for distinguishing negative and positive SLN, and AUC of 0.905 for discriminating low and heavy metastatic burden of SLN. The study also identified two explainable import regions, the tumor boundary and the intratumoral low-echo area, with the gradient-weighted class activation mapping (Grad-CAM).⁴³ Similarly, Liu et al. developed a nomogram with DL features, hand-crafted features, and four clinical parameters, for the preoperative evaluation of ALN metastasis status, with AUCs of 0.91–0.95 in three external validation cohorts.⁴⁴ In conclusion, clinical-radiomic nomograms are effective tools for predicting the SLN or ALN metastasis status, thereby aiding in clinical decision-making.

Prediction of response to neoadjuvant therapy

In the context of NAT as a common treatment of preoperative patients with locally advanced breast cancer, radiomic features derived primarily from MRI are extensively researched as potential biomarkers to predict treatment response before initiating NAT. For instance, a model based on features extracted from pretreatment mpMRI for predicting pathological complete response (pCR) to neoadjuvant chemotherapy (NAC) yielded a maximum AUC of 0.79, while subgrouping by molecular subtypes in advance can further enhance the AUC.⁹⁵ Interestingly, specific imaging patterns related to the NAC response within each molecular subtype were identified. In HR+/HER2− breast cancers, non-pCR tumors exhibit elevated peritumoral heterogeneity during initial contrast enhancement, while TNBC or HER2+ tumors can be distinguished by a speckled enhancement pattern within the peritumoral region of non-pCR tumors.⁹⁶ Besides MRI, PET/CT-based radiomics can also predict pCR to NAC, alone or combined with other modalities or clinical information. Antunovic et al. developed four nomograms combining clinical data, biological characteristics, and PET/CT-based radiomic features and observed a slight of improvement of the AUC (0.71–0.73) when combining SUV_max and TLG in the model.⁹⁸

DL radiomics based on pre- and post-treatment US may also predict pCR to NAC, and researchers have found that nomograms combining hand-crafted features, DL features, and clinical parameters yielded better AUC than single features types.¹⁰¹^,¹⁰²^,¹⁰³ Residual cancer burden (RCB) is another index for measuring the effects of NAC typically assessed through the pathology of surgically excised tumors, and Li et al. have developed a multitask AI system with MRI-based hand-crafted and DL features to classify patients into RCB categories 0–II vs. III and RCB categories 0 and I vs. II and III, reaching highest AUCs in external testing sets of 0.94 and 0.92, respectively.¹⁰⁴ In conclusion, features derived from multiple dimensions and regions, combined with pre-subtyping, collectively contribute to the development of models for predicting the response to NAT.

Prediction of survival and recurrence

Radiomics can also forecast the prognosis of patients with breast cancer by predicting crucial indices such as overall survival (OS), disease-free survival (DFS), recurrence-free survival (RFS), and metastasis-free survival (MFS). For instance, Wang et al. constructed a radiomics score based on MRI and demonstrated its significant association to DFS (p = 0.041). Besides, the study revealed a link between radiomics and computational histopathology, as the immunophenotype and immune cell composition differed among the radiomics score groups.¹⁰⁶ Li et al. reported significant associations between MRI-based radiomic signatures and three widely accepted multigene assay-derived recurrence scores, namely, MammaPrint, Oncotype DX, and PAM50 (r = 0.5–0.56, p < 0.0001).¹¹² Jiang et al. successfully identified a radiomic feature based on DCE-MRI that captured the peritumoral heterogeneity of TNBC as a prognostic factor of RFS (p = 0.01) and OS (p = 0.004).⁸⁸ A previously introduced study for predicting ALN metastasis also established a clinical-radiomic nomogram yielding AUCs of 0.89, 0.91, and 0.90 for 1-year, 2-year, and 3-year DFS prediction, respectively.⁸⁹ Interestingly, Arefan et al. focused on the combination of background parenchymal enhancement score of DCE-MRI and radiomic features for predicting the recurrence risk of breast cancer, which proved to help discover patients with a high Oncotype DX recurrence score with an AUC of 0.79.¹⁰⁷ Research on predicting prognosis using radiogenomic methods will be discussed in the following section. In conclusion, radiomics-based models are crucial and highly effective for predicting the prognosis of patients with breast cancer.

Utilizing radiomics to bridge the gap between phenotypic and microscopic scale information

Beyond predicting clinicopathological indices, radiomics also has the potential to uncover the molecular or biological underpinnings of breast cancer. Radiogenomics initially bridged the gap between phenotypic and microscopic features by exploring the correlation between image phenotypes and genetic profiles, as thoroughly introduced and discussed by Pinker et al., Liu et al., and Wu et al.¹⁰^,¹¹³^,¹¹⁴ We now introduce the concept of radio-multi-omics to further delve into the molecular or biological bases of breast cancer, by exploring the associations between radiomics and multi-omics fields such as genomics, transcriptomics, proteomics, metabolomics, epigenomics, and pathomics. Research exploring the molecular events in breast cancer through radiomics is summarized in Table 2.

Table 2.

Summary of radiomic studies predicting breast cancer molecular events

Molecular event	Number of patients	Modalities	Radiomic features	Modeling algorithms	Feature correlation	Prognosis correlation	Reference
TP53 mutation	Radiomics: N = 94	DCE-MRI	18 related to TP53 mutation	SVM	AUC = 0.85, acc = 0.82, coefficient = 1.74	N.A.	Sun et al.¹¹⁵
TP53 mutation	Radiomics: N = 94	DCE-MRI	1 related to TP53 mutation: T2-wavelet-LHH-GLCM-joint energy	LR, LDA	LR: coefficient = −5.12; LDA: AUC = 0.75, acc = 0.85, coefficient = −4.49	N.A.	Sun et al.¹¹⁵
PD-L1 status	Radiomics: N = 62	DCE-MRI	3 related to PD-L1 status, including variance, run length variance, and large zone low gray level emphasis	Decision tree	AUC = 0.90, sen = 0.90, spe = 0.85, acc = 0.88	N.A.	Lo Gullo et al.¹¹⁶
PD-L1 status	Radiomics: N = 196	DCE-MRI	Intratumoral: 13; Peritumoral: 15; Combined: 18	MRMR, LASSO	Intratumoral: AUC = 0.82; Peritumoral: AUC = 0.85; Combined: AUC = 0.85	N.A.	Wu et al.¹¹⁷
HRD	Radiomics: N = 118; Multi-omics: N = 343	T1-weighted DCE-MRI	3, including the mean value of the first-order minimum features after wavelet-LLL (intratumoral), the kurtosis of the texture GLSZM large area high gray level emphasis feature after wavelet-LHH (peritumoral), and the difference in the first post-enhanced phase and the plain scan value of the wavelet value of the first-order maximum features after wavelet-HHH (peritumoral)	SVM, LASSO, LR, GSEA	SVM: AUC = 0.74, sen = 0.57, spe = 0.824; LR: AUC = 0.70, sen = 0.57, spe = 0.88	N.A.	Su et al.¹¹⁸
Expression of immune-related genes (e.g., STAT1, CXCL9)	Radiomics: N = 10; Transcriptomics: N = 353	3D T1-weighted fat-saturated gradient-echo sequences	Heterogeneous enhancement trait	GSEA, ImageneDx	N.A.	N.A.	Yamamoto et al.¹¹⁹
MammaPrint signature; serum fibroblast, wound healing, hypoxia metagene signatures	Radiomics: N = 10; Transcriptomics: N = 353	3D T1-weighted fat-saturated gradient-echo sequences	11 traits, including retroareolar location, absence of architectural distortion, absence of T2 intrinsic signal strength, etc.	GSEA, ImageneDx	FDR <0.25	N.A.	Yamamoto et al.¹¹⁹
LncRNAs	Radiomics: N = 112; Transcriptomics: N = 58	DCE-MRI	MFS correlated: enhancing rim fraction (ERF)	Correlation analysis	lncRNA HOTAIR and ERF score: r = 0.64, p = 0.003	MFS, ERF score: HR = 6.06, (95% CI: 1.39–26.5), p = 0.017	Yamamoto et al.¹²⁰
A 73-gene signature (mainly TNF signaling pathway)	Radiomics: N = 138; Transcriptomics: N = 150	DCE-MRI	Parenchymal: 10, including dissimilarity, energy, entropy, correlation, etc.	Pearson linear correlation, WGCNA	A 73-gene signature and feature “correlation”: r(2) = 0.87, adjusted r(2) = 0.38	RFS, feature “correlation”: HR = 2.60, (95% CI: 1.43–6.15), p = 0.024	Wu et al.¹²¹
Multigene signatures	Radiomics: N = 61; Transcriptomics: N = 87	DCE-MRI	14, including volume, median value, compactness, maximum probability, etc.; 3 subregion phenotypes, plasma input, fast-flow kinetics, and slow-flow kinetics	CAM, elastic net, KEGG enrichment	A 38-gene signature and tumor volume: r(2) = 0.82; A 43-gene signature and fast-flow kinetics: r(2) = 0.81; A 57-gene signature and slow-flow kinetics: r(2) = 0.90	RFS, feature “median value of precontrast series”: p = 0.0018; RFS, feature “median value of early postcontrast minus precontrast series”: p = 0.0036; RFS, feature “tumor volume”: p = 0.0032	Fan et al.¹²²
Risk genes, gene signatures, and biological pathway activities	Radiomics: N = 137; Transcriptomics: N = 110	T1-weighted DCE-MRI	Hand-crafted features: 36; DL features	LASSO, LME, DA, ssGSEA	1,774 DL features significantly associated with 213 risk genes (p < 0.05); 1,739 DL features significantly associated with 166 KEGG pathways (p < 0.05); 2,028 DL features significantly associated with 381 (213 risk genes, 2 gene signatures, and 166 biological pathways) genomic features	N.A.	Liu et al.¹²³

Open in a new tab

DCE-MRI, dynamic contrast-enhanced magnetic resonance imaging; mpMRI, multiparametric magnetic resonance imaging; lncRNA, long noncoding RNA; DL, deep learning; GLCM, gray-level co-occurrence matrix; GLSZM, gray-level size zone matrix; GLRLM, gray-level run length matrix; SVM, support vector machine; LR, logistic regression; LDA, linear discriminant analysis; MRMR, minimal redundancy maximum relevance; LASSO, least absolute shrinkage and selection operator; GSEA, gene set enrichment analysis; ssGSEA, single sample gene set enrichment analysis; KEGG, Kyoto Encyclopedia of Genes and Genomes; WGCNA, weighted gene coexpression network analysis; cGAN, conditional generative adversarial network; MLP, multilayer neural network; CAM, convex analysis of mixtures; DA, denoising autoencoder; AUC, the area under the curve; sen, sensitivity; spe, specificity; acc, accuracy; N.A., not available; MFS, metastasis-free survival; RFS, recurrence-free survival; HR, hazard ratio; CI, confidence interval.

Predicting molecular events

Apart from traditional ER, PR, HER2, or Ki-67 expression, radiomics helps uncover various specific molecular events related to prognosis and treatment efficacy, including gene mutations and the expression patterns of particular oncogenes or immune factors, thereby aiding clinical decisions in breast cancer treatment. By identifying highly correlated radiomic features, radiomics facilitates the prediction of these events.¹¹⁵^,¹¹⁸ Furthermore, the predictive efficiency of these features for specific molecular events is extensively studied, similar to predicting status of the subtyping-associated molecular events discussed earlier. In the field of correlation with gene mutations, radiomic analysis has been shown to predict the mutations in TP53, as the most frequently mutated gene in breast cancer.¹²⁴ Sun et al. identified radiomic features with multiple models reflecting TP53 mutations in patients with TNBC, among which the SVM model attained the highest AUC (0.85) and accuracy (0.82), and the top 3 ranking features were T1-square-first-order skewness (coefficient: 1.735), T2-wavelet-LHH-GLCM-joint energy (coefficient: −0.654), and T2-wavelet-LHH-GLCM-inverse difference moment (coefficient: −0.634).¹¹⁵ Radiomics has also enabled the non-invasive evaluation of PD-L1 expression status in tumor to select patients who may benefit from anti-PD-1/PD-L1 treatment. Lo Gullo et al. identified DCE-MRI-based radiomic features with significant differences between PD-L1-positive and PD-L1-negative patients and further established a PD-L1 expression level prediction model utilizing 3 MRI parameters, reaching sensitivity of 90.7%, specificity of 85.1%, and accuracy of 88.2%.¹¹⁶ Wu et al. constructed 3 radiomic signatures based on the intratumoral, peritumoral, and combined intra- and peritumoral regions for the preoperative evaluation of PD-L1 expression in breast cancer, reaching AUCs of 0.853, 0.816, and 0.846, respectively.¹¹⁷ Homologous recombination deficiency (HRD), a key damage of DNA double-strand damage repair, is proved fragile to PARPi treatment in TNBC.¹²⁵ Su et al. established SVM and LR radiomics models for predicting the HRD status, reaching AUCs of 0.739 and 0.695, a sensitivity of 0.571 and 0.571, and a specificity of 0.824 and 0.882.¹¹⁸ In conclusion, various known specific prognostic or treatment effect-related molecular events can be predicted by identifying radiomic features with high correlations.

Moreover, novel prognostic or treatment effect-related molecular events or signatures can be discovered based on radio-multi-omics analysis. Yamamoto et al. initially performed radiogenomic association map analysis to discover associations between MRI phenotypes and the expression pattern of genes highly correlated with breast cancer; specifically, a heterogeneous enhancement trait correlated with the high expression levels of immune function genes in the interferon subtype cluster, such as STAT1 and CXCL9. Associations between radiomic features and various prognostic molecular signatures such as the serum fibroblast and wound healing signatures and the hypoxia metagene signature were also validated in the same study.¹¹⁹ In another study, Yamamoto et al. discovered multiscale relationships between radiomic features based on DCE-MRI, long noncoding RNA (lncRNA) expression, and metastasis. Eight lncRNAs were identified for a strong relation with high enhancing rim fraction by radiogenomic analysis and were proved to be predictors of poor MFS in patients with breast cancer.¹²⁰ Wu et al. discovered a 73-gene signature significantly associated with heterogeneous MRI enhancement patterns of the tumor-adjacent parenchyma (r(2) = 0.873), especially the tumor necrosis signaling pathway, with log rank p = 0.00058 for RFS and log rank p = 0.0026 for OS.¹²¹ Fan et al. discovered a correlation between the maximum probability feature from the fast-flow kinetics subregion and the RAS signaling pathway in breast cancer (corrected p value = 0.0044), while a 43-gene signature associated with the maximum probability feature showed significant associations with RFS (p = 0.035) and OS (p = 0.027).¹²² The DL features extracted from output of the last encode hidden layer of a DA model have been shown to correlate with the risk genes, gene signatures, and biological pathway activities (adjusted p < 0.05).¹²³ In conclusion, radio-multi-omics enables a better understanding of the internal relationship between radiomic features and molecular events and helps to predict clinical outcomes or treatment sensitivity (e.g., PARPi), hence providing evidence for precise treatment plan decisions.

Uncovering biological alterations

In addition to various molecular events characterized by genome alterations or the expression status of single molecules, other biological alterations related to prognosis or treatment sensitivity, for instance, ITH, immune activation/suppression, and metabolic dysregulation, can also be predicted through the identification of related radiomic features. ITH is considered to be significantly correlated with poor prognosis or therapeutic effects in breast cancer, yet its traditional assessment depends on invasive transcriptome profiling or histopathology slide examination.¹²⁶ Recently, the use of radiomic models to predict ITH in breast cancer has gained widespread acceptance, with key studies summarized in Table S1. To noninvasively assess the ITH, Su et al. developed an MRI-based method for quantitative measuring imaging ITH, which was further proven to be associated with traditional genomic ITH and digital pathology-defined cellular morphological ITH. The study also delved into metabolomic and transcriptomic analysis and identified activated oncogenic pathways and metabolic dysregulation in radiomics-defined high ITH tumors and discovered ferroptosis as a potential therapeutic target.¹²⁷ From another aspect, Fan et al. quantified cellular tumor-stroma heterogeneity (TSH) by analyzing differences in cell subpopulations between tumor and stroma using transcriptomic data, identifying cytotoxic lymphocytes and fibroblasts as key predictors of TSH score. They then developed a predictive model for TSH score using six radiomic features—four from the tumor and two from the stroma. The model demonstrated prognostic significance, showing a positive correlation with OS (p = 0.038) and RFS (p = 0.0003).¹²⁸ Shi et al. utilized habitat imaging for illustrating intratumoral ecological diversity (iTED) by extracting radiomic features from intratumoral subregions. They built a model based on iTED scores, conventional radiomic scores generated with features extracted from whole-tumor ROIs, and clinicopathological indices for predicting pCR to NAC, reaching AUCs of 0.83–0.87 in external test datasets.¹²⁹ Other radiomic studies with divergent definitions of ITH also showed a satisfactory prediction ability to predict ITH, see Table S1 for details. In conclusion, despite varying criteria and research methods for ITH, the malignancy and association of ITH with poor prognosis or treatment outcomes are well established. Radiomics plays a crucial role in non-invasively predicting ITH, and when integrated with multi-omics analysis, it can further elucidate the biological foundations of ITH.

Besides ITH, the tumor immune microenvironment (TIME), especially TILs, is also an essential biological alteration related to prognosis and therapeutic effects, especially immunotherapy. The immune cell composition or spatial structure of tumor microenvironment can also be predicted utilizing radiomic methods. For predicting TILs in TNBC, Su et al. developed an MRI-based radiomic model with 3 radiomic features utilizing elastic net and LR, with an AUC of 0.79 in the validation cohort. Transcriptomic analysis revealed that the upregulation of pathways correlated with the immune response and immune modulation in the high TIL-radiomics score group, which confirmed the reliability of this radiomic model.¹³⁰ In HER2+ breast cancer, Braman et al. identified 5 radiomic features from the 0- to 3-mm peritumoral region that were significantly associated with the density of TILs in the surrounding tumor environment.⁹⁷ Intratumoral and peritumoral TILs may be correlated with the risk of breast-conserving surgery (BCS) resection margin positivity. Ma et al. developed an MRI-based immune-radiomic model to infer the distribution patterns of TILs for the preoperative prediction of BCS resection margin positivity, with AUCs ranging from 0.68 to 0.94. The study also revealed that positive surgical margins are associated with increased B cell infiltration in the tumor area and higher levels of B cells, immature dendritic cells, and neutrophil infiltration in the peritumoral area.¹³¹ Tertiary lymphoid structure (TLS) is a B cell-enriched spatial structure that can be observed in the TIME, and a higher TLS composition may suggest better prognosis and immunotherapy response in breast cancer.¹³² Li et al. developed a nomogram combining clinical features and 10 selected radiomic features for evaluating TLSs in breast cancer with a maximum AUC of 0.75, and a higher TLS score suggested better 3-year DFS and 3-year distant metastasis-free survival.¹³³ In conclusion, radiomic studies that reveal ITH as well as immune cell infiltration underscore radiomics as a bridge linking macroscopic phenotypes to their microscopic immunological bases.

Current challenges of applying radiomics to clinical practice

As summarized, while radiomics can assist in clinical classification and prediction, most studies are retrospective and these models are seldom used in actual clinical practice. This indicates that radiomics still faces several challenges, including generalizability, interpretability, and convenience in clinical practice.

Generalizability

The generalizability of radiomic studies, that is, the reproducibility of results in large numbers of samples from external cohorts or open databases, relies on the robustness of radiomic models. The relatively low generalizability in radiomics can be attributed to two main factors. Firstly, imperfections in the models arise from poor methodology, small exploratory cohorts, inadequate quality control during model building, and failure to prevent overfitting. Secondly, variations in imaging equipment or protocols across different medical centers can lead to batch effects. Research has focused on exploring the robustness and generalizability of radiomics models. For example, Robinson et al. developed a two-state method for, respectively, robustness assessment and classification evaluation, for selecting robust texture features across full-field digital mammography unit vendors.¹³⁴ Granzier et al. evaluated the robustness of radiomic features extracted by RadiomiX and Pyradiomics and reported that 41.6% and 32.8% of the features, respectively, were robust and independent of inter-observer manual breast tumor segmentation variability.¹³⁵ A satisfactory database is the key to model establishment and validation, and Lambin et al. summarized the 4 Vs of ideal “big data”: volume, variety, velocity, and veracity.⁴⁵ Data sharing in various countries, regions, and medical centers is essential for “big data” building, and various efforts have been made. For example, CancerLinQ, as an initiative of the American Society of Clinical Oncology and its Institute for Quality, aims at data centralization by collecting, measuring, and reporting patients’ multi-dimensional clinical data.¹³⁶ A recent study on exploring ALN metastasis prediction model is a positive example, for the discovery cohort is relatively large (N = 234), and that the validation is well designed, containing two prospective cohorts (N = 81) and one external cohort (N = 723). In this study, the outcome of the comprehensive validation solidly proved the robustness of this model, with ACCs of 0.93, 0.90, and 0.97 in the primary cohort, external validation cohorts, and prospective validation cohort, respectively.⁹³ In conclusion, although various factors can affect the generalizability of certain radiomic models, it is encouraging that more research is focusing on evaluating and optimizing the robustness of these models to improve their generalizability.

Interpretability

Interpretability, also known as transparency, refers to how understandable the radiomic features or models are for human beings. For interpreting hand-crafted feature selection, tree ensemble machine learning algorithms such as the RF and gradient boosting decision tree can provide each feature’s contribution to the outcome.¹³⁷ Besides, the selection of radiomic features according to relevance or importance also promises their interpretability. However, these algorithms are unable to assess inconsistent feature importance. The Shapley additive explanations (SHAP) method is a game theory-based framework, which is shown to be effective in explaining supervised learning models, with the ability to measure the contributions of radiomic features to the increase or decrease in the probability of a single output. Besides hand-crafted features, SHAP also enables the explanation of DL features with DeepExplainer, which calculates the SHAP values for each feature based on the gradient information from neural network-based models.¹³⁸ SHAP has been employed in radiomics studies of breast cancer for explaining both hand-crafted and DL features.⁵⁵^,¹³⁹ Other methods for interpreting DL features include the class activation map (CAM) and CAM-based approaches, such as Grad-CAM and adaptive learning-based CAM, which were proposed for explaining decisions made by a variety of CNN-based models by tracking the area of interest.¹⁴⁰^,¹⁴¹ Additionally, Barnett et al. developed an interpretable CNN algorithm that distinguishes between benign and malignant breast lesions by incorporating a case-based reasoning method. This algorithm specifically highlights regions of the image, indicating that it recognizes these areas as similar to prototypical cases encountered previously. It also provides a diagnostic probability score for the image and assesses the likelihood of malignancy.¹⁴² In conclusion, enhancing model interpretability is crucial yet challenging in radiomic research, particularly in DL studies.

Convenience

Apart from the deficiency of models themselves, the limited convenience of implementing multi-functional models is a key obstacle to the ultimate clinical application of radiomics studies, ranging from tumor classification to decision support. Currently, FDA has approved eight CAD systems for breast cancer screening, including Transpara and INSIGHT MMG. These systems generally serve as adjuncts to experienced radiologists.¹⁴³ These CAD systems are designed to be seamlessly integrated into the user interface and recommend suspicious lesions to radiologists for further evaluation and interpretation, making it possible to accurately diagnose breast cancer with high efficiency.¹⁰⁸ Currently, unlike the CAD systems, the interpretable radiomics-based models, both ML-based and DL-based, have not yet been adopted for clinical screening or for other applications. This may be partly attributed to the fact that most developed models are monofunctional, requiring doctors to use separate models for each specific purpose, which reduces efficiency. Therefore, there is an urgent need to develop more convenient and user-friendly models to enhance the acceptance of radiomic models in clinical settings.

Future directions

We propose future directions to enhance the generalizability, interpretability, and convenience of radiomics in breast cancer. These include employing optimized validation methods, such as multicenter and prospective trials, to achieve higher evidence levels; delving deeper into DL radiomics and adopting radio-multi-omics to enhance interpretability; and developing a user-friendly, all-in-one multifunctional system to improve convenience and promote clinical application (Figure 2).

Optimizing validation strategies

Multicenter validation

As discussed earlier, validating the reliability of established radiomic models through external validation using large-scale multicenter cohorts or data from open databases is crucial for demonstrating their generalizability. This process requires the intimate cooperation of medical centers, preferably internationally. Fortunately, recent studies on breast cancer increasingly emphasize robust validation, with examples of studies undergoing external validation shown in Table 1. The previous studies at our center have highlighted the significance of this validation, aiming to expand the application of our model across more medical centers.⁸⁸^,¹²⁷ Looking ahead, we anticipate that more radiomic studies will undergo external multicenter validation.

Prospective clinical trials

Radiomic models developed from retrospective datasets require further validation through prospective clinical trials to attain higher levels of evidence. Therefore, we should extend these trials to encompass all aspects of clinical scenarios. Prospective trials should evaluate both the classification accuracy and the efficiency of therapeutic effect predictions of these models compared to a cohort treated according to gold-standard recommendations. In conclusion, the ultimate goal of developing radiomic models is their integration into clinical decision-making, emphasizing the importance of prospective validation.

Enhancing interpretability

From hand-craft radiomics to DL radiomics

DL radiomics is increasingly being explored by researchers to build prediction models for various applications in breast cancer. DL methods can identify specific and subtle imaging features that surpass human observation and those extracted by traditional ML models. These advances significantly enhance the predictive accuracy for diverse clinical events and represent a significant trend in future developments. While enhancing the interpretability of the “black box” “end-to-end” DL models remains challenging, DL radiomics allows the extraction of DL features and modeling with conventional ML methods for higher interpretability. In brief, we anticipate greater and more transparent use of DL radiomics in future breast cancer clinical scenarios.

From radiogenomics to radio-multi-omics

As mentioned earlier, radiomics acts as a bridge linking macroscopical phenotypes to their biological underpinnings. Consequently, we anticipate that radiomic models will not only assist in clinical decision-making but also delve deeper into exploring the correlations between imaging features and biological events, ultimately leading to the prediction of occurrences of significant events. Looking forward, we envision a shift from radiogenomics to radio-multi-omics, which involves integrating multidimensional data in radiomic studies, including genomics, transcriptomics, proteomics, metabolomics, epigenomics, pathomics, etc. The benefit of utilizing multidimensional data lies in their complementary nature, which provides a comprehensive view of tumor biology, and their inherent correlations that can enhance mutual mapping. In summary, future studies should aim to broaden the understanding of imaging features through comprehensive radio-multi-omics analysis.

Establishing an all-in-one AI-aided radiology-clinic system

Inspired by the successful real-world application of CAD systems in breast cancer screening, we propose a user-friendly all-in-one system that integrates multiple functions from tumor classification to outcome prediction, hoping that the diverse potential clinical applications of radiomics can be realized. The radiomic models of this multifunctional system should be pretrained and proven efficient and robust. Users begin by entering a patient’s unique identifier, such as their hospital number, and then choose the personalized demand. The system then automatically gathers and processes comprehensive clinicopathological data, medical images from various modalities, and multi-omics information. For screening demand, the DL section of the system gives a preliminary diagnosis of the lesion and marks suspicious lesion for radiologists, with approved CAD systems. If the lesion is proved malignant by radiologists, or for patients already diagnosed with breast cancer, comprehensive information will be automatically gathered and processed by the system, including comprehensive clinicopathological data, medical images from various modalities, and multi-omics information. Then, the radiomic section of the system automatically predicts the molecular subtype and lymph node status of the patient. For the patients who will receive NAT, or have already undergone surgery, the system automatically predicts the prognosis and treatment response. Finally, the system synthesizes the outcomes and outputs a personalized and integrated report with detailed explanations for both clinical doctors and patients. We believe that enhancing the system’s user-friendliness will improve its acceptability among medical professionals and patients, potentially facilitating its clinical adoption.

In conclusion, in this review, we summarize the current utilization of radiomics in breast cancer, including the development of classification or prediction models, as well as correlating with or predicting biological events. We also emphasize the essential role of radiomics in clinical practice, where it may support both radiologists and clinicians. Despite existing barriers to clinical application, which suggest further optimization of the current models, we provide several practical recommendations and look forward to the integration of radiomics into clinical practice for breast cancer.

Acknowledgments

This work was supported by grants from the National Natural Science Foundation of China (82373167, 92159301, and 82271957), the Shanghai Key Laboratory of Breast Cancer (12DZ2260100), Shanghai Medical Innovation Research Project (22Y11912700), and Shanghai Anticancer Association EYAS PROJECT (SACA-CY22A05). The funders had no role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript. We acknowledge the assistance of Dongdong Zheng (Department of Ultrasound, Fudan University Shanghai Cancer Center and Department of Oncology, Shanghai Medical College, Fudan University, Shanghai 200032, P.R. China) for providing us with representative medical images. We thank BioRender (biorender.com) for assisting in drawing graphics.

Author contributions

Conceptualization, Y.X., Y.-Z.J., and Z.-M.S.; writing – original draft, Y.-J.Q., G.-H.S., and C.Y.; writing – review and editing, Y.-J.Q., G.-H.S., C.Y., and X.Z.; funding acquisition, Y.X., Y.-Z.J., and Z.-M.S.; supervision, Y.X., Y.-Z.J., and Z.-M.S.

Declaration of interests

The authors declare no competing interests.

Footnotes

Supplemental information can be found online at https://doi.org/10.1016/j.xcrm.2024.101719.

Contributor Information

Yi Xiao, Email: yixiao11@fudan.edu.cn.

Yi-Zhou Jiang, Email: yizhoujiang@fudan.edu.cn.

Zhi-Ming Shao, Email: zhimingshao@fudan.edu.cn.

Supplemental information

Document S1. Table S1

mmc1.pdf^{(106.7KB, pdf)}

Document S2. Article plus supplemental information

mmc2.pdf^{(3.6MB, pdf)}

References

1.Bray F., Laversanne M., Sung H., Ferlay J., Siegel R.L., Soerjomataram I., Jemal A. Global cancer statistics 2022: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA A Cancer J. Clin. 2024;74:229–263. doi: 10.3322/caac.21834. [DOI] [PubMed] [Google Scholar]
2.Zardavas D., Irrthum A., Swanton C., Piccart M. Clinical management of breast cancer heterogeneity. Nat. Rev. Clin. Oncol. 2015;12:381–394. doi: 10.1038/nrclinonc.2015.73. [DOI] [PubMed] [Google Scholar]
3.Nolan E., Lindeman G.J., Visvader J.E. Deciphering breast cancer: from biology to the clinic. Cell. 2023;186:1708–1728. doi: 10.1016/j.cell.2023.01.040. [DOI] [PubMed] [Google Scholar]
4.Hartmann K., Sadée C.Y., Satwah I., Carrillo-Perez F., Gevaert O. Imaging genomics: data fusion in uncovering disease heritability. Trends Mol. Med. 2023;29:141–151. doi: 10.1016/j.molmed.2022.11.002. [DOI] [PMC free article] [PubMed] [Google Scholar]
5.Lambin P., Rios-Velazquez E., Leijenaar R., Carvalho S., van Stiphout R.G.P.M., Granton P., Zegers C.M.L., Gillies R., Boellard R., Dekker A., Aerts H.J.W.L. Radiomics: extracting more information from medical images using advanced feature analysis. Eur. J. Cancer. 2012;48:441–446. doi: 10.1016/j.ejca.2011.11.036. [DOI] [PMC free article] [PubMed] [Google Scholar]
6.Gillies R.J., Kinahan P.E., Hricak H. Radiomics: Images Are More than Pictures, They Are Data. Radiology. 2016;278:563–577. doi: 10.1148/radiol.2015151169. [DOI] [PMC free article] [PubMed] [Google Scholar]
7.Bi W.L., Hosny A., Schabath M.B., Giger M.L., Birkbak N.J., Mehrtash A., Allison T., Arnaout O., Abbosh C., Dunn I.F., et al. Artificial intelligence in cancer imaging: Clinical challenges and applications. CA A Cancer J. Clin. 2019;69:127–157. doi: 10.3322/caac.21552. [DOI] [PMC free article] [PubMed] [Google Scholar]
8.Zhang Y.P., Zhang X.Y., Cheng Y.T., Li B., Teng X.Z., Zhang J., Lam S., Zhou T., Ma Z.R., Sheng J.B., et al. Artificial intelligence-driven radiomics study in cancer: the role of feature engineering and modeling. Mil. Med. Res. 2023;10:22. doi: 10.1186/s40779-023-00458-8. [DOI] [PMC free article] [PubMed] [Google Scholar]
9.Shao J., Ma J., Zhang Q., Li W., Wang C. Predicting gene mutation status via artificial intelligence technologies based on multimodal integration (MMI) to advance precision oncology. Semin. Cancer Biol. 2023;91:1–15. doi: 10.1016/j.semcancer.2023.02.006. [DOI] [PubMed] [Google Scholar]
10.Pinker K., Chin J., Melsaether A.N., Morris E.A., Moy L. Precision Medicine and Radiogenomics in Breast Cancer: New Approaches toward Diagnosis and Treatment. Radiology. 2018;287:732–747. doi: 10.1148/radiol.2018172171. [DOI] [PubMed] [Google Scholar]
11.Huang E.P., O'Connor J.P.B., McShane L.M., Giger M.L., Lambin P., Kinahan P.E., Siegel E.L., Shankar L.K. Criteria for the translation of radiomics into clinically useful tests. Nat. Rev. Clin. Oncol. 2023;20:69–82. doi: 10.1038/s41571-022-00707-0. [DOI] [PMC free article] [PubMed] [Google Scholar]
12.Siu A.L., U.S. Preventive Services Task Force Screening for Breast Cancer: U.S. Preventive Services Task Force Recommendation Statement. Ann. Intern. Med. 2016;164:279–296. doi: 10.7326/m15-2886. [DOI] [PubMed] [Google Scholar]
13.Farkas A.H., Nattinger A.B. Breast Cancer Screening and Prevention. Ann. Intern. Med. 2023;176:ITC161–ITC176. doi: 10.7326/aitc202311210. Itc161-itc176. [DOI] [PubMed] [Google Scholar]
14.Clift A.K., Dodwell D., Lord S., Petrou S., Brady S.M., Collins G.S., Hippisley-Cox J. The current status of risk-stratified breast screening. Br. J. Cancer. 2022;126:533–550. doi: 10.1038/s41416-021-01550-3. [DOI] [PMC free article] [PubMed] [Google Scholar]
15.Goldhirsch A., Wood W.C., Coates A.S., Gelber R.D., Thürlimann B., Senn H.J., Panel members Strategies for subtypes--dealing with the diversity of breast cancer: highlights of the St. Gallen International Expert Consensus on the Primary Therapy of Early Breast Cancer 2011. Ann. Oncol. 2011;22:1736–1747. doi: 10.1093/annonc/mdr304. [DOI] [PMC free article] [PubMed] [Google Scholar]
16.Prat A., Pineda E., Adamo B., Galván P., Fernández A., Gaba L., Díez M., Viladot M., Arance A., Muñoz M. Clinical implications of the intrinsic molecular subtypes of breast cancer. Breast. 2015;24:S26–S35. doi: 10.1016/j.breast.2015.07.008. [DOI] [PubMed] [Google Scholar]
17.Leon-Ferre R.A., Goetz M.P. Advances in systemic therapies for triple negative breast cancer. Bmj. 2023;381 doi: 10.1136/bmj-2022-071674. [DOI] [PubMed] [Google Scholar]
18.Oeffinger K.C., Fontham E.T.H., Etzioni R., Herzig A., Michaelson J.S., Shih Y.C.T., Walter L.C., Church T.R., Flowers C.R., LaMonte S.J., et al. Breast Cancer Screening for Women at Average Risk: 2015 Guideline Update From the American Cancer Society. JAMA. 2015;314:1599–1614. doi: 10.1001/jama.2015.12783. [DOI] [PMC free article] [PubMed] [Google Scholar]
19.Schünemann H.J., Lerda D., Quinn C., Follmann M., Alonso-Coello P., Rossi P.G., Lebeau A., Nyström L., Broeders M., Ioannidou-Mouzaka L., et al. Breast Cancer Screening and Diagnosis: A Synopsis of the European Breast Guidelines. Ann. Intern. Med. 2020;172:46–56. doi: 10.7326/m19-2125. [DOI] [PubMed] [Google Scholar]
20.Filipits M., Rudas M., Jakesz R., Dubsky P., Fitzal F., Singer C.F., Dietze O., Greil R., Jelen A., Sevelda P., et al. A new molecular predictor of distant recurrence in ER-positive, HER2-negative breast cancer adds independent information to conventional clinical risk factors. Clin. Cancer Res. 2011;17:6012–6020. doi: 10.1158/1078-0432.Ccr-11-0926. [DOI] [PubMed] [Google Scholar]
21.Cobleigh M.A., Tabesh B., Bitterman P., Baker J., Cronin M., Liu M.L., Borchik R., Mosquera J.M., Walker M.G., Shak S. Tumor gene expression and prognosis in breast cancer patients with 10 or more positive lymph nodes. Clin. Cancer Res. 2005;11:8623–8631. doi: 10.1158/1078-0432.Ccr-05-0735. [DOI] [PubMed] [Google Scholar]
22.Parker J.S., Mullins M., Cheang M.C.U., Leung S., Voduc D., Vickery T., Davies S., Fauron C., He X., Hu Z., et al. Supervised risk predictor of breast cancer based on intrinsic subtypes. J. Clin. Oncol. 2009;27:1160–1167. doi: 10.1200/jco.2008.18.1370. [DOI] [PMC free article] [PubMed] [Google Scholar]
23.van de Vijver M.J., He Y.D., van't Veer L.J., Dai H., Hart A.A.M., Voskuil D.W., Schreiber G.J., Peterse J.L., Roberts C., Marton M.J., et al. A gene-expression signature as a predictor of survival in breast cancer. N. Engl. J. Med. 2002;347:1999–2009. doi: 10.1056/NEJMoa021967. [DOI] [PubMed] [Google Scholar]
24.Sharma P., Stecklein S.R., Yoder R., Staley J.M., Schwensen K., O'Dea A., Nye L., Satelli D., Crane G., Madan R., et al. Clinical and Biomarker Findings of Neoadjuvant Pembrolizumab and Carboplatin Plus Docetaxel in Triple-Negative Breast Cancer: NeoPACT Phase 2 Clinical Trial. JAMA Oncol. 2024;10:227–235. doi: 10.1001/jamaoncol.2023.5033. [DOI] [PMC free article] [PubMed] [Google Scholar]
25.Liu Z., Wang S., Dong D., Wei J., Fang C., Zhou X., Sun K., Li L., Li B., Wang M., Tian J. The Applications of Radiomics in Precision Diagnosis and Treatment of Oncology: Opportunities and Challenges. Theranostics. 2019;9:1303–1322. doi: 10.7150/thno.30309. [DOI] [PMC free article] [PubMed] [Google Scholar]
26.Conti A., Duggento A., Indovina I., Guerrisi M., Toschi N. Radiomics in breast cancer classification and prediction. Semin. Cancer Biol. 2021;72:238–250. doi: 10.1016/j.semcancer.2020.04.002. [DOI] [PubMed] [Google Scholar]
27.Houssami N., Zackrisson S., Blazek K., Hunter K., Bernardi D., Lång K., Hofvind S. Meta-analysis of prospective studies evaluating breast cancer detection and interval cancer rates for digital breast tomosynthesis versus mammography population screening. Eur. J. Cancer. 2021;148:14–23. doi: 10.1016/j.ejca.2021.01.035. [DOI] [PubMed] [Google Scholar]
28.Ghaderi K.F., Phillips J., Perry H., Lotfi P., Mehta T.S. Contrast-enhanced Mammography: Current Applications and Future Directions. Radiographics. 2019;39:1907–1920. doi: 10.1148/rg.2019190079. [DOI] [PubMed] [Google Scholar]
29.Scheel J.R., Lee J.M., Sprague B.L., Lee C.I., Lehman C.D. Screening ultrasound as an adjunct to mammography in women with mammographically dense breasts. Am. J. Obstet. Gynecol. 2015;212:9–17. doi: 10.1016/j.ajog.2014.06.048. [DOI] [PMC free article] [PubMed] [Google Scholar]
30.Sigrist R.M.S., Liau J., Kaffas A.E., Chammas M.C., Willmann J.K. Ultrasound Elastography: Review of Techniques and Clinical Applications. Theranostics. 2017;7:1303–1329. doi: 10.7150/thno.18650. [DOI] [PMC free article] [PubMed] [Google Scholar]
31.Zanotel M., Bednarova I., Londero V., Linda A., Lorenzon M., Girometti R., Zuiani C. Automated breast ultrasound: basic principles and emerging clinical applications. Radiol. Med. 2018;123:1–12. doi: 10.1007/s11547-017-0805-z. [DOI] [PubMed] [Google Scholar]
32.Satake H., Ishigaki S., Ito R., Naganawa S. Radiomics in breast MRI: current progress toward clinical application in the era of artificial intelligence. Radiol. Med. 2022;127:39–56. doi: 10.1007/s11547-021-01423-y. [DOI] [PubMed] [Google Scholar]
33.Urso L., Manco L., Castello A., Evangelista L., Guidi G., Castellani M., Florimonte L., Cittanti C., Turra A., Panareo S. PET-Derived Radiomics and Artificial Intelligence in Breast Cancer: A Systematic Review. Int. J. Mol. Sci. 2022;23 doi: 10.3390/ijms232113409. [DOI] [PMC free article] [PubMed] [Google Scholar]
34.Lai S., Liang F., Zhang W., Zhao Y., Li J., Zhao Y., Xu Y., Ding W., Zhan J., Zhen X., Yang R. Evaluation of molecular receptors status in breast cancer using an mpMRI-based feature fusion radiomics model: mimicking radiologists' diagnosis. Front. Oncol. 2023;13 doi: 10.3389/fonc.2023.1219071. [DOI] [PMC free article] [PubMed] [Google Scholar]
35.Li Y., Han D., Shen C., Duan X. Construction of a comprehensive predictive model for axillary lymph node metastasis in breast cancer: a retrospective study. BMC Cancer. 2023;23:1028. doi: 10.1186/s12885-023-11498-7. [DOI] [PMC free article] [PubMed] [Google Scholar]
36.Mayerhoefer M.E., Materka A., Langs G., Häggström I., Szczypiński P., Gibbs P., Cook G. Introduction to Radiomics. J. Nucl. Med. 2020;61:488–495. doi: 10.2967/jnumed.118.222893. [DOI] [PMC free article] [PubMed] [Google Scholar]
37.Abbasian Ardakani A., Bureau N.J., Ciaccio E.J., Acharya U.R. Interpretation of radiomics features-A pictorial review. Comput. Methods Progr. Biomed. 2022;215 doi: 10.1016/j.cmpb.2021.106609. [DOI] [PubMed] [Google Scholar]
38.Haralick R.M., Shanmugam K., Dinstein I. Textural features for image classification. IEEE Trans. Syst. Man Cybern. 1973;SMC-3:610–621. [Google Scholar]
39.Chen X., Wang X., Zhang K., Fung K.M., Thai T.C., Moore K., Mannel R.S., Liu H., Zheng B., Qiu Y. Recent advances and clinical applications of deep learning in medical image analysis. Med. Image Anal. 2022;79 doi: 10.1016/j.media.2022.102444. [DOI] [PMC free article] [PubMed] [Google Scholar]
40.LeCun Y., Bengio Y., Hinton G. Deep learning. Nature. 2015;521:436–444. doi: 10.1038/nature14539. [DOI] [PubMed] [Google Scholar]
41.Yang X., Wu L., Zhao K., Ye W., Liu W., Wang Y., Li J., Li H., Huang X., Zhang W., et al. Evaluation of human epidermal growth factor receptor 2 status of breast cancer using preoperative multidetector computed tomography with deep learning and handcrafted radiomics features. Chin. J. Cancer Res. 2020;32:175–185. doi: 10.21147/j.issn.1000-9604.2020.02.05. [DOI] [PMC free article] [PubMed] [Google Scholar]
42.Huang Y., Yao Z., Li L., Mao R., Huang W., Hu Z., Hu Y., Wang Y., Guo R., Tang X., et al. Deep learning radiopathomics based on preoperative US images and biopsy whole slide images can distinguish between luminal and non-luminal tumors in early-stage breast cancers. EBioMedicine. 2023;94 doi: 10.1016/j.ebiom.2023.104706. [DOI] [PMC free article] [PubMed] [Google Scholar]
43.Zheng X., Yao Z., Huang Y., Yu Y., Wang Y., Liu Y., Mao R., Li F., Xiao Y., Wang Y., et al. Deep learning radiomics can predict axillary lymph node status in early-stage breast cancer. Nat. Commun. 2020;11:1236. doi: 10.1038/s41467-020-15027-z. [DOI] [PMC free article] [PubMed] [Google Scholar]
44.Liu H., Zou L., Xu N., Shen H., Zhang Y., Wan P., Wen B., Zhang X., He Y., Gui L., Kong W. Deep learning radiomics based prediction of axillary lymph node metastasis in breast cancer. NPJ Breast Cancer. 2024;10:22. doi: 10.1038/s41523-024-00628-4. [DOI] [PMC free article] [PubMed] [Google Scholar]
45.Lambin P., Leijenaar R.T.H., Deist T.M., Peerlings J., de Jong E.E.C., van Timmeren J., Sanduleanu S., Larue R.T.H.M., Even A.J.G., Jochems A., et al. Radiomics: the bridge between medical imaging and personalized medicine. Nat. Rev. Clin. Oncol. 2017;14:749–762. doi: 10.1038/nrclinonc.2017.141. [DOI] [PubMed] [Google Scholar]
46.Ma Q., Li Z., Li W., Chen Q., Liu X., Feng W., Lei J. MRI radiomics for the preoperative evaluation of lymphovascular invasion in breast cancer: A meta-analysis. Eur. J. Radiol. 2023;168 doi: 10.1016/j.ejrad.2023.111127. [DOI] [PubMed] [Google Scholar]
47.Dercle L., McGale J., Sun S., Marabelle A., Yeh R., Deutsch E., Mokrane F.Z., Farwell M., Ammari S., Schoder H., et al. Artificial intelligence and radiomics: fundamentals, applications, and challenges in immunotherapy. J. Immunother. Cancer. 2022;10 doi: 10.1136/jitc-2022-005292. [DOI] [PMC free article] [PubMed] [Google Scholar]
48.Zhao B. Understanding Sources of Variation to Improve the Reproducibility of Radiomics. Front. Oncol. 2021;11 doi: 10.3389/fonc.2021.633176. [DOI] [PMC free article] [PubMed] [Google Scholar]
49.Clarke L.P., Nordstrom R.J., Zhang H., Tandon P., Zhang Y., Redmond G., Farahani K., Kelloff G., Henderson L., Shankar L., et al. The Quantitative Imaging Network: NCI's Historical Perspective and Planned Goals. Transl. Oncol. 2014;7:1–4. doi: 10.1593/tlo.13832. [DOI] [PMC free article] [PubMed] [Google Scholar]
50.Kalpathy-Cramer J., Freymann J.B., Kirby J.S., Kinahan P.E., Prior F.W. Quantitative Imaging Network: Data Sharing and Competitive AlgorithmValidation Leveraging The Cancer Imaging Archive. Transl. Oncol. 2014;7:147–152. doi: 10.1593/tlo.13862. [DOI] [PMC free article] [PubMed] [Google Scholar]
51.Shukla-Dave A., Obuchowski N.A., Chenevert T.L., Jambawalikar S., Schwartz L.H., Malyarenko D., Huang W., Noworolski S.M., Young R.J., Shiroishi M.S., et al. Quantitative imaging biomarkers alliance (QIBA) recommendations for improved precision of DWI and DCE-MRI derived biomarkers in multicenter oncology trials. J. Magn. Reson. Imag. 2019;49:e101–e121. doi: 10.1002/jmri.26518. [DOI] [PMC free article] [PubMed] [Google Scholar]
52.Limkin E.J., Sun R., Dercle L., Zacharaki E.I., Robert C., Reuzé S., Schernberg A., Paragios N., Deutsch E., Ferté C. Promises and challenges for the implementation of computational medical imaging (radiomics) in oncology. Ann. Oncol. 2017;28:1191–1206. doi: 10.1093/annonc/mdx034. [DOI] [PubMed] [Google Scholar]
53.Parmar C., Rios Velazquez E., Leijenaar R., Jermoumi M., Carvalho S., Mak R.H., Mitra S., Shankar B.U., Kikinis R., Haibe-Kains B., et al. Robust Radiomics feature quantification using semiautomatic volumetric segmentation. PLoS One. 2014;9 doi: 10.1371/journal.pone.0102107. [DOI] [PMC free article] [PubMed] [Google Scholar]
54.Zhang J., Cui Z., Shi Z., Jiang Y., Zhang Z., Dai X., Yang Z., Gu Y., Zhou L., Han C., et al. A robust and efficient AI assistant for breast tumor segmentation from DCE-MRI via a spatial-temporal framework. Patterns (N Y) 2023;4 doi: 10.1016/j.patter.2023.100826. [DOI] [PMC free article] [PubMed] [Google Scholar]
55.Beuque M.P.L., Lobbes M.B.I., van Wijk Y., Widaatalla Y., Primakov S., Majer M., Balleyguier C., Woodruff H.C., Lambin P. Combining Deep Learning and Handcrafted Radiomics for Classification of Suspicious Lesions on Contrast-enhanced Mammograms. Radiology. 2023;307 doi: 10.1148/radiol.221843. [DOI] [PubMed] [Google Scholar]
56.Zhang X., Su G.H., Chen Y., Gu Y.J., You C. Decoding Intratumoral Heterogeneity: Clinical Potential of Habitat Imaging based on Radiomics. Radiology. 2023;309 doi: 10.1148/radiol.232047. [DOI] [PubMed] [Google Scholar]
57.Torre V., Poggio T.A. On edge detection. IEEE Trans. Pattern Anal. Mach. Intell. 1986;8:147–163. doi: 10.1109/tpami.1986.4767769. [DOI] [PubMed] [Google Scholar]
58.Farge M. Wavelet transforms and their applications to turbulence. Annu. Rev. Fluid Mech. 1992;24:395–457. [Google Scholar]
59.Zwanenburg A., Vallières M., Abdalah M.A., Aerts H.J.W.L., Andrearczyk V., Apte A., Ashrafinia S., Bakas S., Beukinga R.J., Boellaard R., et al. The Image Biomarker Standardization Initiative: Standardized Quantitative Radiomics for High-Throughput Image-based Phenotyping. Radiology. 2020;295:328–338. doi: 10.1148/radiol.2020191145. [DOI] [PMC free article] [PubMed] [Google Scholar]
60.Kovalev V.A., Kruggel F., Gertz H.J., von Cramon D.Y. Three-dimensional texture analysis of MRI brain datasets. IEEE Trans. Med. Imag. 2001;20:424–433. doi: 10.1109/42.925295. [DOI] [PubMed] [Google Scholar]
61.Zhou J., Luo L.Y., Dou Q., Chen H., Chen C., Li G.J., Jiang Z.F., Heng P.A. Weakly supervised 3D deep learning for breast cancer classification and localization of the lesions in MR images. J. Magn. Reson. Imag. 2019;50:1144–1151. doi: 10.1002/jmri.26721. [DOI] [PubMed] [Google Scholar]
62.Bagherzadeh-Khiabani F., Ramezankhani A., Azizi F., Hadaegh F., Steyerberg E.W., Khalili D. A tutorial on variable selection for clinical prediction models: feature selection methods in data mining could improve the results. J. Clin. Epidemiol. 2016;71:76–85. doi: 10.1016/j.jclinepi.2015.10.002. [DOI] [PubMed] [Google Scholar]
63.Herrero Vicent C., Tudela X., Moreno Ruiz P., Pedralva V., Jiménez Pastor A., Ahicart D., Rubio Novella S., Meneu I., Montes Albuixech Á., Santamaria M.Á., et al. Machine Learning Models and Multiparametric Magnetic Resonance Imaging for the Prediction of Pathologic Response to Neoadjuvant Chemotherapy in Breast Cancer. Cancers. 2022;14 doi: 10.3390/cancers14143508. [DOI] [PMC free article] [PubMed] [Google Scholar]
64.Breiman L. Random forests. Mach. Learn. 2001;45:5–32. [Google Scholar]
65.Hearst M.A., Dumais S.T., Osuna E., Platt J., Scholkopf B. Support vector machines. IEEE Intell. Syst. Their Appl. 1998;13:18–28. [Google Scholar]
66.Tibshirani R. Regression shrinkage and selection via the lasso. J. Roy. Stat. Soc. B Stat. Methodol. 1996;58:267–288. [Google Scholar]
67.Kanungo T., Mount D.M., Netanyahu N.S., Piatko C.D., Silverman R., Wu A.Y. An efficient k-means clustering algorithm: Analysis and implementation. IEEE Trans. Pattern Anal. Mach. Intell. 2002;24:881–892. [Google Scholar]
68.Monti S., Tamayo P., Mesirov J., Golub T. Consensus clustering: a resampling-based method for class discovery and visualization of gene expression microarray data. Mach. Learn. 2003;52:91–118. [Google Scholar]
69.Bezdek J.C., Ehrlich R., Full W. FCM: The fuzzy c-means clustering algorithm. Comput. Geosci. 1984;10:191–203. [Google Scholar]
70.Goodfellow I., Pouget-Abadie J., Mirza M., Xu B., Warde-Farley D., Ozair S., Courville A., Bengio Y. Generative adversarial nets. Adv. Neural Inf. Process. Syst. 2014;27 [Google Scholar]
71.Vincent P., Larochelle H., Bengio Y., Manzagol P.-A. 2008. Extracting and Composing Robust Features with Denoising Autoencoders; pp. 1096–1103. [Google Scholar]
72.Joachims T. 1999. Transductive Inference for Text Classification Using Support Vector Machines; pp. 200–209. [Google Scholar]
73.Salimans T., Goodfellow I., Zaremba W., Cheung V., Radford A., Chen X. Improved techniques for training gans. Adv. Neural Inf. Process. Syst. 2016;29 [Google Scholar]
74.Halligan S., Menu Y., Mallett S. Why did European Radiology reject my radiomic biomarker paper? How to correctly evaluate imaging biomarkers in a clinical setting. Eur. Radiol. 2021;31:9361–9368. doi: 10.1007/s00330-021-07971-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
75.Kocak B., Durmaz E.S., Erdim C., Ates E., Kaya O.K., Kilickesmez O. Radiomics of Renal Masses: Systematic Review of Reproducibility and Validation Strategies. AJR Am. J. Roentgenol. 2020;214:129–136. doi: 10.2214/ajr.19.21709. [DOI] [PubMed] [Google Scholar]
76.Harrell F.E., Jr., Lee K.L., Califf R.M., Pryor D.B., Rosati R.A. Regression modelling strategies for improved prognostic prediction. Stat. Med. 1984;3:143–152. doi: 10.1002/sim.4780030207. [DOI] [PubMed] [Google Scholar]
77.Heagerty P.J., Lumley T., Pepe M.S. Time-dependent ROC curves for censored survival data and a diagnostic marker. Biometrics. 2000;56:337–344. doi: 10.1111/j.0006-341x.2000.00337.x. [DOI] [PubMed] [Google Scholar]
78.Li H., Mendel K.R., Lan L., Sheth D., Giger M.L. Digital Mammography in Breast Cancer: Additive Value of Radiomics of Breast Parenchyma. Radiology. 2019;291:15–20. doi: 10.1148/radiol.2019181113. [DOI] [PMC free article] [PubMed] [Google Scholar]
79.Drukker K., Giger M.L., Joe B.N., Kerlikowske K., Greenwood H., Drukteinis J.S., Niell B., Fan B., Malkov S., Avila J., et al. Combined Benefit of Quantitative Three-Compartment Breast Image Analysis and Mammography Radiomics in the Classification of Breast Masses in a Clinical Data Set. Radiology. 2019;290:621–628. doi: 10.1148/radiol.2018180608. [DOI] [PMC free article] [PubMed] [Google Scholar]
80.Bickelhaupt S., Jaeger P.F., Laun F.B., Lederer W., Daniel H., Kuder T.A., Wuesthof L., Paech D., Bonekamp D., Radbruch A., et al. Radiomics Based on Adapted Diffusion Kurtosis Imaging Helps to Clarify Most Mammographic Findings Suspicious for Cancer. Radiology. 2018;287:761–770. doi: 10.1148/radiol.2017170273. [DOI] [PubMed] [Google Scholar]
81.Zhang J., Zhan C., Zhang C., Song Y., Yan X., Guo Y., Ai T., Yang G. Fully automatic classification of breast lesions on multi-parameter MRI using a radiomics model with minimal number of stable, interpretable features. Radiol. Med. 2023;128:160–170. doi: 10.1007/s11547-023-01594-w. [DOI] [PubMed] [Google Scholar]
82.Truhn D., Schrading S., Haarburger C., Schneider H., Merhof D., Kuhl C. Radiomic versus Convolutional Neural Networks Analysis for Classification of Contrast-enhancing Lesions at Multiparametric Breast MRI. Radiology. 2019;290:290–297. doi: 10.1148/radiol.2018181352. [DOI] [PubMed] [Google Scholar]
83.Li H., Zhu Y., Burnside E.S., Huang E., Drukker K., Hoadley K.A., Fan C., Conzen S.D., Zuley M., Net J.M., et al. Quantitative MRI radiomics in the prediction of molecular classifications of breast cancer subtypes in the TCGA/TCIA data set. NPJ Breast Cancer. 2016;2 doi: 10.1038/npjbcancer.2016.12. [DOI] [PMC free article] [PubMed] [Google Scholar]
84.Leithner D., Horvat J.V., Marino M.A., Bernard-Davila B., Jochelson M.S., Ochoa-Albiztegui R.E., Martinez D.F., Morris E.A., Thakur S., Pinker K. Radiomic signatures with contrast-enhanced magnetic resonance imaging for the assessment of breast cancer receptor status and molecular subtypes: initial results. Breast Cancer Res. 2019;21:106. doi: 10.1186/s13058-019-1187-z. [DOI] [PMC free article] [PubMed] [Google Scholar]
85.Petrillo A., Fusco R., Petrosino T., Vallone P., Granata V., Rubulotta M.R., Pariante P., Raiano N., Scognamiglio G., Fanizzi A., et al. A multicentric study of radiomics and artificial intelligence analysis on contrast-enhanced mammography to identify different histotypes of breast cancer. Radiol. Med. 2024;129:864–878. doi: 10.1007/s11547-024-01817-8. [DOI] [PubMed] [Google Scholar]
86.Ji Y., Whitney H.M., Li H., Liu P., Giger M.L., Zhang X. Differences in Molecular Subtype Reference Standards Impact AI-based Breast Cancer Classification with Dynamic Contrast-enhanced MRI. Radiology. 2023;307 doi: 10.1148/radiol.220984. [DOI] [PMC free article] [PubMed] [Google Scholar]
87.Umutlu L., Kirchner J., Bruckmann N.M., Morawitz J., Antoch G., Ingenwerth M., Bittner A.K., Hoffmann O., Haubold J., Grueneisen J., et al. Multiparametric Integrated (18)F-FDG PET/MRI-Based Radiomics for Breast Cancer Phenotyping and Tumor Decoding. Cancers. 2021;13 doi: 10.3390/cancers13122928. [DOI] [PMC free article] [PubMed] [Google Scholar]
88.Jiang L., You C., Xiao Y., Wang H., Su G.H., Xia B.Q., Zheng R.C., Zhang D.D., Jiang Y.Z., Gu Y.J., Shao Z.M. Radiogenomic analysis reveals tumor heterogeneity of triple-negative breast cancer. Cell Rep. Med. 2022;3 doi: 10.1016/j.xcrm.2022.100694. [DOI] [PMC free article] [PubMed] [Google Scholar]
89.Yu Y., Tan Y., Xie C., Hu Q., Ouyang J., Chen Y., Gu Y., Li A., Lu N., He Z., et al. Development and Validation of a Preoperative Magnetic Resonance Imaging Radiomics-Based Signature to Predict Axillary Lymph Node Metastasis and Disease-Free Survival in Patients With Early-Stage Breast Cancer. JAMA Netw. Open. 2020;3 doi: 10.1001/jamanetworkopen.2020.28086. [DOI] [PMC free article] [PubMed] [Google Scholar]
90.Han L., Zhu Y., Liu Z., Yu T., He C., Jiang W., Kan Y., Dong D., Tian J., Luo Y. Radiomic nomogram for prediction of axillary lymph node metastasis in breast cancer. Eur. Radiol. 2019;29:3820–3829. doi: 10.1007/s00330-018-5981-2. [DOI] [PubMed] [Google Scholar]
91.Zhan C., Hu Y., Wang X., Liu H., Xia L., Ai T. Prediction of Axillary Lymph Node Metastasis in Breast Cancer using Intra-peritumoral Textural Transition Analysis based on Dynamic Contrast-enhanced Magnetic Resonance Imaging. Acad. Radiol. 2022;29 doi: 10.1016/j.acra.2021.02.008. S107-s115. [DOI] [PubMed] [Google Scholar]
92.Gao Y., Luo Y., Zhao C., Xiao M., Ma L., Li W., Qin J., Zhu Q., Jiang Y. Nomogram based on radiomics analysis of primary breast cancer ultrasound images: prediction of axillary lymph node tumor burden in patients. Eur. Radiol. 2021;31:928–937. doi: 10.1007/s00330-020-07181-1. [DOI] [PubMed] [Google Scholar]
93.Zhu T., Huang Y.H., Li W., Zhang Y.M., Lin Y.Y., Cheng M.Y., Wu Z.Y., Ye G.L., Lin Y., Wang K. Multifactor artificial intelligence model assists axillary lymph node surgery in breast cancer after neoadjuvant chemotherapy: multicenter retrospective cohort study. Int. J. Surg. 2023;109:3383–3394. doi: 10.1097/js9.0000000000000621. [DOI] [PMC free article] [PubMed] [Google Scholar]
94.Qu L., Mei X., Yi Z., Zou Q., Zhou Q., Zhang D., Zhou M., Pei L., Long Q., Meng J., et al. An unsupervised learning model based on CT radiomics features accurately predicts axillary lymph node metastasis in breast cancer patients-diagnostic study. Int. J. Surg. 2024 doi: 10.1097/js9.0000000000001778. [DOI] [PMC free article] [PubMed] [Google Scholar]
95.Liu Z., Li Z., Qu J., Zhang R., Zhou X., Li L., Sun K., Tang Z., Jiang H., Li H., et al. Radiomics of Multiparametric MRI for Pretreatment Prediction of Pathologic Complete Response to Neoadjuvant Chemotherapy in Breast Cancer: A Multicenter Study. Clin. Cancer Res. 2019;25:3538–3547. doi: 10.1158/1078-0432.Ccr-18-3190. [DOI] [PubMed] [Google Scholar]
96.Braman N.M., Etesami M., Prasanna P., Dubchuk C., Gilmore H., Tiwari P., Plecha D., Madabhushi A. Intratumoral and peritumoral radiomics for the pretreatment prediction of pathological complete response to neoadjuvant chemotherapy based on breast DCE-MRI. Breast Cancer Res. 2017;19:57. doi: 10.1186/s13058-017-0846-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
97.Braman N., Prasanna P., Whitney J., Singh S., Beig N., Etesami M., Bates D.D.B., Gallagher K., Bloch B.N., Vulchi M., et al. Association of Peritumoral Radiomics With Tumor Biology and Pathologic Response to Preoperative Targeted Therapy for HER2 (ERBB2)-Positive Breast Cancer. JAMA Netw. Open. 2019;2 doi: 10.1001/jamanetworkopen.2019.2561. [DOI] [PMC free article] [PubMed] [Google Scholar]
98.Antunovic L., De Sanctis R., Cozzi L., Kirienko M., Sagona A., Torrisi R., Tinterri C., Santoro A., Chiti A., Zelic R., Sollini M. PET/CT radiomics in breast cancer: promising tool for prediction of pathological response to neoadjuvant chemotherapy. Eur. J. Nucl. Med. Mol. Imag. 2019;46:1468–1477. doi: 10.1007/s00259-019-04313-8. [DOI] [PubMed] [Google Scholar]
99.Zhang H., Cao W., Liu L., Meng Z., Sun N., Meng Y., Fei J. Noninvasive prediction of node-positive breast cancer response to presurgical neoadjuvant chemotherapy therapy based on machine learning of axillary lymph node ultrasound. J. Transl. Med. 2023;21:337. doi: 10.1186/s12967-023-04201-8. [DOI] [PMC free article] [PubMed] [Google Scholar]
100.Mao N., Shi Y., Lian C., Wang Z., Zhang K., Xie H., Zhang H., Chen Q., Cheng G., Xu C., Dai Y. Intratumoral and peritumoral radiomics for preoperative prediction of neoadjuvant chemotherapy effect in breast cancer based on contrast-enhanced spectral mammography. Eur. Radiol. 2022;32:3207–3219. doi: 10.1007/s00330-021-08414-7. [DOI] [PubMed] [Google Scholar]
101.Jiang M., Li C.L., Luo X.M., Chuan Z.R., Lv W.Z., Li X., Cui X.W., Dietrich C.F. Ultrasound-based deep learning radiomics in the assessment of pathological complete response to neoadjuvant chemotherapy in locally advanced breast cancer. Eur. J. Cancer. 2021;147:95–105. doi: 10.1016/j.ejca.2021.01.028. [DOI] [PubMed] [Google Scholar]
102.Huang Y., Zhu T., Zhang X., Li W., Zheng X., Cheng M., Ji F., Zhang L., Yang C., Wu Z., et al. Longitudinal MRI-based fusion novel model predicts pathological complete response in breast cancer treated with neoadjuvant chemotherapy: a multicenter, retrospective study. EClinicalMedicine. 2023;58 doi: 10.1016/j.eclinm.2023.101899. [DOI] [PMC free article] [PubMed] [Google Scholar]
103.Fu Y., Lei Y.T., Huang Y.H., Mei F., Wang S., Yan K., Wang Y.H., Ma Y.H., Cui L.G. Longitudinal ultrasound-based AI model predicts axillary lymph node response to neoadjuvant chemotherapy in breast cancer: a multicenter study. Eur. Radiol. 2024 doi: 10.1007/s00330-024-10786-5. [DOI] [PMC free article] [PubMed] [Google Scholar]
104.Li W., Huang Y.H., Zhu T., Zhang Y.M., Zheng X.X., Zhang T.F., Lin Y.Y., Wu Z.Y., Liu Z.Y., Lin Y., et al. Noninvasive Artificial Intelligence System for Early Predicting Residual Cancer Burden during Neoadjuvant Chemotherapy in Breast Cancer. Ann. Surg. 2024 doi: 10.1097/sla.0000000000006279. [DOI] [PMC free article] [PubMed] [Google Scholar]
105.Park H., Lim Y., Ko E.S., Cho H.H., Lee J.E., Han B.K., Ko E.Y., Choi J.S., Park K.W. Radiomics Signature on Magnetic Resonance Imaging: Association with Disease-Free Survival in Patients with Invasive Breast Cancer. Clin. Cancer Res. 2018;24:4705–4714. doi: 10.1158/1078-0432.Ccr-17-3783. [DOI] [PubMed] [Google Scholar]
106.Wang X., Xie T., Luo J., Zhou Z., Yu X., Guo X. Radiomics predicts the prognosis of patients with locally advanced breast cancer by reflecting the heterogeneity of tumor cells and the tumor microenvironment. Breast Cancer Res. 2022;24:20. doi: 10.1186/s13058-022-01516-0. [DOI] [PMC free article] [PubMed] [Google Scholar]
107.Arefan D., Zuley M.L., Berg W.A., Yang L., Sumkin J.H., Wu S. Assessment of Background Parenchymal Enhancement at Dynamic Contrast-enhanced MRI in Predicting Breast Cancer Recurrence Risk. Radiology. 2024;310 doi: 10.1148/radiol.230269. [DOI] [PMC free article] [PubMed] [Google Scholar]
108.Geras K.J., Mann R.M., Moy L. Artificial Intelligence for Mammography and Digital Breast Tomosynthesis: Current Concepts and Future Perspectives. Radiology. 2019;293:246–259. doi: 10.1148/radiol.2019182627. [DOI] [PMC free article] [PubMed] [Google Scholar]
109.Díaz O., Rodríguez-Ruíz A., Sechopoulos I. Artificial Intelligence for breast cancer detection: Technology, challenges, and prospects. Eur. J. Radiol. 2024;175 doi: 10.1016/j.ejrad.2024.111457. [DOI] [PubMed] [Google Scholar]
110.Mudigonda N.R., Rangayyan R.M., Desautels J.E. Gradient and texture analysis for the classification of mammographic masses. IEEE Trans. Med. Imag. 2000;19:1032–1043. doi: 10.1109/42.887618. [DOI] [PubMed] [Google Scholar]
111.Mazurowski M.A., Zhang J., Grimm L.J., Yoon S.C., Silber J.I. Radiogenomic analysis of breast cancer: luminal B molecular subtype is associated with enhancement dynamics at MR imaging. Radiology. 2014;273:365–372. doi: 10.1148/radiol.14132641. [DOI] [PubMed] [Google Scholar]
112.Li H., Zhu Y., Burnside E.S., Drukker K., Hoadley K.A., Fan C., Conzen S.D., Whitman G.J., Sutton E.J., Net J.M., et al. MR Imaging Radiomics Signatures for Predicting the Risk of Breast Cancer Recurrence as Given by Research Versions of MammaPrint, Oncotype DX, and PAM50 Gene Assays. Radiology. 2016;281:382–391. doi: 10.1148/radiol.2016152110. [DOI] [PMC free article] [PubMed] [Google Scholar]
113.Wu J., Mayer A.T., Li R. Integrated imaging and molecular analysis to decipher tumor microenvironment in the era of immunotherapy. Semin. Cancer Biol. 2022;84:310–328. doi: 10.1016/j.semcancer.2020.12.005. [DOI] [PMC free article] [PubMed] [Google Scholar]
114.Liu Z., Duan T., Zhang Y., Weng S., Xu H., Ren Y., Zhang Z., Han X. Radiogenomics: a key component of precision cancer medicine. Br. J. Cancer. 2023;129:741–753. doi: 10.1038/s41416-023-02317-8. [DOI] [PMC free article] [PubMed] [Google Scholar]
115.Sun K., Zhu H., Chai W., Yan F. Multimodality MRI radiomics analysis of TP53 mutations in triple negative breast cancer. Front. Oncol. 2023;13 doi: 10.3389/fonc.2023.1153261. [DOI] [PMC free article] [PubMed] [Google Scholar]
116.Lo Gullo R., Wen H., Reiner J.S., Hoda R., Sevilimedu V., Martinez D.F., Thakur S.B., Jochelson M.S., Gibbs P., Pinker K. Assessing PD-L1 Expression Status Using Radiomic Features from Contrast-Enhanced Breast MRI in Breast Cancer Patients: Initial Results. Cancers. 2021;13 doi: 10.3390/cancers13246273. [DOI] [PMC free article] [PubMed] [Google Scholar]
117.Wu Z., Lin Q., Wang H., Chen J., Wang G., Fu G., Li L., Bian T. Intratumoral and Peritumoral Radiomics Based on Preoperative MRI for Evaluation of Programmed Cell Death Ligand-1 Expression in Breast Cancer. J. Magn. Reson. Imag. 2024;60:588–599. doi: 10.1002/jmri.29109. [DOI] [PubMed] [Google Scholar]
118.Su G.H., Jiang L., Xiao Y., Zheng R.C., Wang H., Jiang Y.Z., Peng W.J., Shao Z.M., Gu Y.J., You C. A Multiomics Signature Highlights Alterations Underlying Homologous Recombination Deficiency in Triple-Negative Breast Cancer. Ann. Surg Oncol. 2022;29:7165–7175. doi: 10.1245/s10434-022-11958-7. [DOI] [PubMed] [Google Scholar]
119.Yamamoto S., Maki D.D., Korn R.L., Kuo M.D. Radiogenomic analysis of breast cancer using MRI: a preliminary study to define the landscape. AJR Am. J. Roentgenol. 2012;199:654–663. doi: 10.2214/ajr.11.7824. [DOI] [PubMed] [Google Scholar]
120.Yamamoto S., Han W., Kim Y., Du L., Jamshidi N., Huang D., Kim J.H., Kuo M.D. Breast Cancer: Radiogenomic Biomarker Reveals Associations among Dynamic Contrast-enhanced MR Imaging, Long Noncoding RNA, and Metastasis. Radiology. 2015;275:384–392. doi: 10.1148/radiol.15142698. [DOI] [PubMed] [Google Scholar]
121.Wu J., Li B., Sun X., Cao G., Rubin D.L., Napel S., Ikeda D.M., Kurian A.W., Li R. Heterogeneous Enhancement Patterns of Tumor-adjacent Parenchyma at MR Imaging Are Associated with Dysregulated Signaling Pathways and Poor Survival in Breast Cancer. Radiology. 2017;285:401–413. doi: 10.1148/radiol.2017162823. [DOI] [PMC free article] [PubMed] [Google Scholar]
122.Fan M., Xia P., Liu B., Zhang L., Wang Y., Gao X., Li L. Tumour heterogeneity revealed by unsupervised decomposition of dynamic contrast-enhanced magnetic resonance imaging is associated with underlying gene expression patterns and poor survival in breast cancer patients. Breast Cancer Res. 2019;21:112. doi: 10.1186/s13058-019-1199-8. [DOI] [PMC free article] [PubMed] [Google Scholar]
123.Liu Q., Hu P. Radiogenomic association of deep MR imaging features with genomic profiles and clinical characteristics in breast cancer. Biomark. Res. 2023;11:9. doi: 10.1186/s40364-023-00455-y. [DOI] [PMC free article] [PubMed] [Google Scholar]
124.Shahbandi A., Nguyen H.D., Jackson J.G. TP53 Mutations and Outcomes in Breast Cancer: Reading beyond the Headlines. Trends Cancer. 2020;6:98–110. doi: 10.1016/j.trecan.2020.01.007. [DOI] [PMC free article] [PubMed] [Google Scholar]
125.Chopra N., Tovey H., Pearson A., Cutts R., Toms C., Proszek P., Hubank M., Dowsett M., Dodson A., Daley F., et al. Homologous recombination DNA repair deficiency and PARP inhibition activity in primary triple negative breast cancer. Nat. Commun. 2020;11:2662. doi: 10.1038/s41467-020-16142-7. [DOI] [PMC free article] [PubMed] [Google Scholar]
126.Li L., Patil D., Petruncio G., Harnden K.K., Somasekharan J.V., Paige M., Wang L.V., Salvador-Morales C. Integration of Multitargeted Polymer-Based Contrast Agents with Photoacoustic Computed Tomography: An Imaging Technique to Visualize Breast Cancer Intratumor Heterogeneity. ACS Nano. 2021;15:2413–2427. doi: 10.1021/acsnano.0c05893. [DOI] [PMC free article] [PubMed] [Google Scholar]
127.Su G.H., Xiao Y., You C., Zheng R.C., Zhao S., Sun S.Y., Zhou J.Y., Lin L.Y., Wang H., Shao Z.M., et al. Radiogenomic-based multiomic analysis reveals imaging intratumor heterogeneity phenotypes and therapeutic targets. Sci. Adv. 2023;9 doi: 10.1126/sciadv.adf0837. [DOI] [PMC free article] [PubMed] [Google Scholar]
128.Fan M., Wang K., Zhang Y., Ge Y., Lü Z., Li L. Radiogenomic analysis of cellular tumor-stroma heterogeneity as a prognostic predictor in breast cancer. J. Transl. Med. 2023;21:851. doi: 10.1186/s12967-023-04748-6. [DOI] [PMC free article] [PubMed] [Google Scholar]
129.Shi Z., Huang X., Cheng Z., Xu Z., Lin H., Liu C., Chen X., Liu C., Liang C., Lu C., et al. MRI-based Quantification of Intratumoral Heterogeneity for Predicting Treatment Response to Neoadjuvant Chemotherapy in Breast Cancer. Radiology. 2023;308 doi: 10.1148/radiol.222830. [DOI] [PubMed] [Google Scholar]
130.Su G.H., Xiao Y., Jiang L., Zheng R.C., Wang H., Chen Y., Gu Y.J., You C., Shao Z.M. Radiomics features for assessing tumor-infiltrating lymphocytes correlate with molecular traits of triple-negative breast cancer. J. Transl. Med. 2022;20:471. doi: 10.1186/s12967-022-03688-x. [DOI] [PMC free article] [PubMed] [Google Scholar]
131.Ma J., Chen K., Li S., Zhu L., Yu Y., Li J., Ma J., Ouyang J., Wu Z., Tan Y., et al. MRI-based radiomic models to predict surgical margin status and infer tumor immune microenvironment in breast cancer patients with breast-conserving surgery: a multicenter validation study. Eur. Radiol. 2024;34:1774–1789. doi: 10.1007/s00330-023-10144-x. [DOI] [PubMed] [Google Scholar]
132.Wang Q., Sun K., Liu R., Song Y., Lv Y., Bi P., Yang F., Li S., Zhao J., Li X., et al. Single-cell transcriptome sequencing of B-cell heterogeneity and tertiary lymphoid structure predicts breast cancer prognosis and neoadjuvant therapy efficacy. Clin. Transl. Med. 2023;13 doi: 10.1002/ctm2.1346. [DOI] [PMC free article] [PubMed] [Google Scholar]
133.Li K., Ji J., Li S., Yang M., Che Y., Xu Z., Zhang Y., Wang M., Fang Z., Luo L., et al. Analysis of the Correlation and Prognostic Significance of Tertiary Lymphoid Structures in Breast Cancer: A Radiomics-Clinical Integration Approach. J. Magn. Reson. Imag. 2024;59:1206–1217. doi: 10.1002/jmri.28900. [DOI] [PubMed] [Google Scholar]
134.Robinson K., Li H., Lan L., Schacht D., Giger M. Radiomics robustness assessment and classification evaluation: A two-stage method demonstrated on multivendor FFDM. Med. Phys. 2019;46:2145–2156. doi: 10.1002/mp.13455. [DOI] [PMC free article] [PubMed] [Google Scholar]
135.Granzier R.W.Y., Verbakel N.M.H., Ibrahim A., van Timmeren J.E., van Nijnatten T.J.A., Leijenaar R.T.H., Lobbes M.B.I., Smidt M.L., Woodruff H.C. MRI-based radiomics in breast cancer: feature robustness with respect to inter-observer segmentation variability. Sci. Rep. 2020;10 doi: 10.1038/s41598-020-70940-z. [DOI] [PMC free article] [PubMed] [Google Scholar]
136.Schilsky R.L., Michels D.L., Kearbey A.H., Yu P.P., Hudis C.A. Building a rapid learning health care system for oncology: the regulatory framework of CancerLinQ. J. Clin. Oncol. 2014;32:2373–2379. doi: 10.1200/jco.2014.56.2124. [DOI] [PubMed] [Google Scholar]
137.Friedman J.H. Greedy function approximation: a gradient boosting machine. Ann. Stat. 2001;29:1189–1232. [Google Scholar]
138.Lundberg S.M., Lee S.-I. Advances in neural information processing systems 30. 2017. A unified approach to interpreting model predictions. [Google Scholar]
139.Huang Y., Wang X., Cao Y., Li M., Li L., Chen H., Tang S., Lan X., Jiang F., Zhang J. Multiparametric MRI model to predict molecular subtypes of breast cancer using Shapley additive explanations interpretability analysis. Diagn. Interv. Imaging. 2024;105:191–205. doi: 10.1016/j.diii.2024.01.004. [DOI] [PubMed] [Google Scholar]
140.Selvaraju R.R., Cogswell M., Das A., Vedantam R., Parikh D., Batra D. 2017. Grad-cam: Visual Explanations from Deep Networks via Gradient-Based Localization; pp. 618–626. [Google Scholar]
141.Iqbal S., Qureshi A.N., Alhussein M., Aurangzeb K., Anwar M.S. AD-CAM: Enhancing Interpretability of Convolutional Neural Networks with a Lightweight Framework - From Black Box to Glass Box. IEEE J. Biomed. Health Inform. 2023 doi: 10.1109/jbhi.2023.3329231. [DOI] [PubMed] [Google Scholar]
142.Barnett A.J., Schwartz F.R., Tao C., Chen C., Ren Y., Lo J.Y., Rudin C. A case-based interpretable deep learning model for classification of mass lesions in digital mammography. Nat. Mach. Intell. 2021;3:1061–1070. [Google Scholar]
143.Loizidou K., Elia R., Pitris C. Computer-aided breast cancer detection and classification in mammography: A comprehensive review. Comput. Biol. Med. 2023;153 doi: 10.1016/j.compbiomed.2023.106554. [DOI] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Document S1. Table S1

mmc1.pdf^{(106.7KB, pdf)}

Document S2. Article plus supplemental information

mmc2.pdf^{(3.6MB, pdf)}

[bib1] 1.Bray F., Laversanne M., Sung H., Ferlay J., Siegel R.L., Soerjomataram I., Jemal A. Global cancer statistics 2022: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA A Cancer J. Clin. 2024;74:229–263. doi: 10.3322/caac.21834. [DOI] [PubMed] [Google Scholar]

[bib2] 2.Zardavas D., Irrthum A., Swanton C., Piccart M. Clinical management of breast cancer heterogeneity. Nat. Rev. Clin. Oncol. 2015;12:381–394. doi: 10.1038/nrclinonc.2015.73. [DOI] [PubMed] [Google Scholar]

[bib3] 3.Nolan E., Lindeman G.J., Visvader J.E. Deciphering breast cancer: from biology to the clinic. Cell. 2023;186:1708–1728. doi: 10.1016/j.cell.2023.01.040. [DOI] [PubMed] [Google Scholar]

[bib4] 4.Hartmann K., Sadée C.Y., Satwah I., Carrillo-Perez F., Gevaert O. Imaging genomics: data fusion in uncovering disease heritability. Trends Mol. Med. 2023;29:141–151. doi: 10.1016/j.molmed.2022.11.002. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib5] 5.Lambin P., Rios-Velazquez E., Leijenaar R., Carvalho S., van Stiphout R.G.P.M., Granton P., Zegers C.M.L., Gillies R., Boellard R., Dekker A., Aerts H.J.W.L. Radiomics: extracting more information from medical images using advanced feature analysis. Eur. J. Cancer. 2012;48:441–446. doi: 10.1016/j.ejca.2011.11.036. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib6] 6.Gillies R.J., Kinahan P.E., Hricak H. Radiomics: Images Are More than Pictures, They Are Data. Radiology. 2016;278:563–577. doi: 10.1148/radiol.2015151169. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib7] 7.Bi W.L., Hosny A., Schabath M.B., Giger M.L., Birkbak N.J., Mehrtash A., Allison T., Arnaout O., Abbosh C., Dunn I.F., et al. Artificial intelligence in cancer imaging: Clinical challenges and applications. CA A Cancer J. Clin. 2019;69:127–157. doi: 10.3322/caac.21552. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib8] 8.Zhang Y.P., Zhang X.Y., Cheng Y.T., Li B., Teng X.Z., Zhang J., Lam S., Zhou T., Ma Z.R., Sheng J.B., et al. Artificial intelligence-driven radiomics study in cancer: the role of feature engineering and modeling. Mil. Med. Res. 2023;10:22. doi: 10.1186/s40779-023-00458-8. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib9] 9.Shao J., Ma J., Zhang Q., Li W., Wang C. Predicting gene mutation status via artificial intelligence technologies based on multimodal integration (MMI) to advance precision oncology. Semin. Cancer Biol. 2023;91:1–15. doi: 10.1016/j.semcancer.2023.02.006. [DOI] [PubMed] [Google Scholar]

[bib10] 10.Pinker K., Chin J., Melsaether A.N., Morris E.A., Moy L. Precision Medicine and Radiogenomics in Breast Cancer: New Approaches toward Diagnosis and Treatment. Radiology. 2018;287:732–747. doi: 10.1148/radiol.2018172171. [DOI] [PubMed] [Google Scholar]

[bib11] 11.Huang E.P., O'Connor J.P.B., McShane L.M., Giger M.L., Lambin P., Kinahan P.E., Siegel E.L., Shankar L.K. Criteria for the translation of radiomics into clinically useful tests. Nat. Rev. Clin. Oncol. 2023;20:69–82. doi: 10.1038/s41571-022-00707-0. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib12] 12.Siu A.L., U.S. Preventive Services Task Force Screening for Breast Cancer: U.S. Preventive Services Task Force Recommendation Statement. Ann. Intern. Med. 2016;164:279–296. doi: 10.7326/m15-2886. [DOI] [PubMed] [Google Scholar]

[bib13] 13.Farkas A.H., Nattinger A.B. Breast Cancer Screening and Prevention. Ann. Intern. Med. 2023;176:ITC161–ITC176. doi: 10.7326/aitc202311210. Itc161-itc176. [DOI] [PubMed] [Google Scholar]

[bib14] 14.Clift A.K., Dodwell D., Lord S., Petrou S., Brady S.M., Collins G.S., Hippisley-Cox J. The current status of risk-stratified breast screening. Br. J. Cancer. 2022;126:533–550. doi: 10.1038/s41416-021-01550-3. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib15] 15.Goldhirsch A., Wood W.C., Coates A.S., Gelber R.D., Thürlimann B., Senn H.J., Panel members Strategies for subtypes--dealing with the diversity of breast cancer: highlights of the St. Gallen International Expert Consensus on the Primary Therapy of Early Breast Cancer 2011. Ann. Oncol. 2011;22:1736–1747. doi: 10.1093/annonc/mdr304. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib16] 16.Prat A., Pineda E., Adamo B., Galván P., Fernández A., Gaba L., Díez M., Viladot M., Arance A., Muñoz M. Clinical implications of the intrinsic molecular subtypes of breast cancer. Breast. 2015;24:S26–S35. doi: 10.1016/j.breast.2015.07.008. [DOI] [PubMed] [Google Scholar]

[bib17] 17.Leon-Ferre R.A., Goetz M.P. Advances in systemic therapies for triple negative breast cancer. Bmj. 2023;381 doi: 10.1136/bmj-2022-071674. [DOI] [PubMed] [Google Scholar]

[bib18] 18.Oeffinger K.C., Fontham E.T.H., Etzioni R., Herzig A., Michaelson J.S., Shih Y.C.T., Walter L.C., Church T.R., Flowers C.R., LaMonte S.J., et al. Breast Cancer Screening for Women at Average Risk: 2015 Guideline Update From the American Cancer Society. JAMA. 2015;314:1599–1614. doi: 10.1001/jama.2015.12783. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib19] 19.Schünemann H.J., Lerda D., Quinn C., Follmann M., Alonso-Coello P., Rossi P.G., Lebeau A., Nyström L., Broeders M., Ioannidou-Mouzaka L., et al. Breast Cancer Screening and Diagnosis: A Synopsis of the European Breast Guidelines. Ann. Intern. Med. 2020;172:46–56. doi: 10.7326/m19-2125. [DOI] [PubMed] [Google Scholar]

[bib20] 20.Filipits M., Rudas M., Jakesz R., Dubsky P., Fitzal F., Singer C.F., Dietze O., Greil R., Jelen A., Sevelda P., et al. A new molecular predictor of distant recurrence in ER-positive, HER2-negative breast cancer adds independent information to conventional clinical risk factors. Clin. Cancer Res. 2011;17:6012–6020. doi: 10.1158/1078-0432.Ccr-11-0926. [DOI] [PubMed] [Google Scholar]

[bib21] 21.Cobleigh M.A., Tabesh B., Bitterman P., Baker J., Cronin M., Liu M.L., Borchik R., Mosquera J.M., Walker M.G., Shak S. Tumor gene expression and prognosis in breast cancer patients with 10 or more positive lymph nodes. Clin. Cancer Res. 2005;11:8623–8631. doi: 10.1158/1078-0432.Ccr-05-0735. [DOI] [PubMed] [Google Scholar]

[bib22] 22.Parker J.S., Mullins M., Cheang M.C.U., Leung S., Voduc D., Vickery T., Davies S., Fauron C., He X., Hu Z., et al. Supervised risk predictor of breast cancer based on intrinsic subtypes. J. Clin. Oncol. 2009;27:1160–1167. doi: 10.1200/jco.2008.18.1370. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib23] 23.van de Vijver M.J., He Y.D., van't Veer L.J., Dai H., Hart A.A.M., Voskuil D.W., Schreiber G.J., Peterse J.L., Roberts C., Marton M.J., et al. A gene-expression signature as a predictor of survival in breast cancer. N. Engl. J. Med. 2002;347:1999–2009. doi: 10.1056/NEJMoa021967. [DOI] [PubMed] [Google Scholar]

[bib24] 24.Sharma P., Stecklein S.R., Yoder R., Staley J.M., Schwensen K., O'Dea A., Nye L., Satelli D., Crane G., Madan R., et al. Clinical and Biomarker Findings of Neoadjuvant Pembrolizumab and Carboplatin Plus Docetaxel in Triple-Negative Breast Cancer: NeoPACT Phase 2 Clinical Trial. JAMA Oncol. 2024;10:227–235. doi: 10.1001/jamaoncol.2023.5033. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib25] 25.Liu Z., Wang S., Dong D., Wei J., Fang C., Zhou X., Sun K., Li L., Li B., Wang M., Tian J. The Applications of Radiomics in Precision Diagnosis and Treatment of Oncology: Opportunities and Challenges. Theranostics. 2019;9:1303–1322. doi: 10.7150/thno.30309. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib26] 26.Conti A., Duggento A., Indovina I., Guerrisi M., Toschi N. Radiomics in breast cancer classification and prediction. Semin. Cancer Biol. 2021;72:238–250. doi: 10.1016/j.semcancer.2020.04.002. [DOI] [PubMed] [Google Scholar]

[bib27] 27.Houssami N., Zackrisson S., Blazek K., Hunter K., Bernardi D., Lång K., Hofvind S. Meta-analysis of prospective studies evaluating breast cancer detection and interval cancer rates for digital breast tomosynthesis versus mammography population screening. Eur. J. Cancer. 2021;148:14–23. doi: 10.1016/j.ejca.2021.01.035. [DOI] [PubMed] [Google Scholar]

[bib28] 28.Ghaderi K.F., Phillips J., Perry H., Lotfi P., Mehta T.S. Contrast-enhanced Mammography: Current Applications and Future Directions. Radiographics. 2019;39:1907–1920. doi: 10.1148/rg.2019190079. [DOI] [PubMed] [Google Scholar]

[bib29] 29.Scheel J.R., Lee J.M., Sprague B.L., Lee C.I., Lehman C.D. Screening ultrasound as an adjunct to mammography in women with mammographically dense breasts. Am. J. Obstet. Gynecol. 2015;212:9–17. doi: 10.1016/j.ajog.2014.06.048. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib30] 30.Sigrist R.M.S., Liau J., Kaffas A.E., Chammas M.C., Willmann J.K. Ultrasound Elastography: Review of Techniques and Clinical Applications. Theranostics. 2017;7:1303–1329. doi: 10.7150/thno.18650. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib31] 31.Zanotel M., Bednarova I., Londero V., Linda A., Lorenzon M., Girometti R., Zuiani C. Automated breast ultrasound: basic principles and emerging clinical applications. Radiol. Med. 2018;123:1–12. doi: 10.1007/s11547-017-0805-z. [DOI] [PubMed] [Google Scholar]

[bib32] 32.Satake H., Ishigaki S., Ito R., Naganawa S. Radiomics in breast MRI: current progress toward clinical application in the era of artificial intelligence. Radiol. Med. 2022;127:39–56. doi: 10.1007/s11547-021-01423-y. [DOI] [PubMed] [Google Scholar]

[bib33] 33.Urso L., Manco L., Castello A., Evangelista L., Guidi G., Castellani M., Florimonte L., Cittanti C., Turra A., Panareo S. PET-Derived Radiomics and Artificial Intelligence in Breast Cancer: A Systematic Review. Int. J. Mol. Sci. 2022;23 doi: 10.3390/ijms232113409. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib34] 34.Lai S., Liang F., Zhang W., Zhao Y., Li J., Zhao Y., Xu Y., Ding W., Zhan J., Zhen X., Yang R. Evaluation of molecular receptors status in breast cancer using an mpMRI-based feature fusion radiomics model: mimicking radiologists' diagnosis. Front. Oncol. 2023;13 doi: 10.3389/fonc.2023.1219071. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib35] 35.Li Y., Han D., Shen C., Duan X. Construction of a comprehensive predictive model for axillary lymph node metastasis in breast cancer: a retrospective study. BMC Cancer. 2023;23:1028. doi: 10.1186/s12885-023-11498-7. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib36] 36.Mayerhoefer M.E., Materka A., Langs G., Häggström I., Szczypiński P., Gibbs P., Cook G. Introduction to Radiomics. J. Nucl. Med. 2020;61:488–495. doi: 10.2967/jnumed.118.222893. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib37] 37.Abbasian Ardakani A., Bureau N.J., Ciaccio E.J., Acharya U.R. Interpretation of radiomics features-A pictorial review. Comput. Methods Progr. Biomed. 2022;215 doi: 10.1016/j.cmpb.2021.106609. [DOI] [PubMed] [Google Scholar]

[bib38] 38.Haralick R.M., Shanmugam K., Dinstein I. Textural features for image classification. IEEE Trans. Syst. Man Cybern. 1973;SMC-3:610–621. [Google Scholar]

[bib39] 39.Chen X., Wang X., Zhang K., Fung K.M., Thai T.C., Moore K., Mannel R.S., Liu H., Zheng B., Qiu Y. Recent advances and clinical applications of deep learning in medical image analysis. Med. Image Anal. 2022;79 doi: 10.1016/j.media.2022.102444. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib40] 40.LeCun Y., Bengio Y., Hinton G. Deep learning. Nature. 2015;521:436–444. doi: 10.1038/nature14539. [DOI] [PubMed] [Google Scholar]

[bib41] 41.Yang X., Wu L., Zhao K., Ye W., Liu W., Wang Y., Li J., Li H., Huang X., Zhang W., et al. Evaluation of human epidermal growth factor receptor 2 status of breast cancer using preoperative multidetector computed tomography with deep learning and handcrafted radiomics features. Chin. J. Cancer Res. 2020;32:175–185. doi: 10.21147/j.issn.1000-9604.2020.02.05. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib42] 42.Huang Y., Yao Z., Li L., Mao R., Huang W., Hu Z., Hu Y., Wang Y., Guo R., Tang X., et al. Deep learning radiopathomics based on preoperative US images and biopsy whole slide images can distinguish between luminal and non-luminal tumors in early-stage breast cancers. EBioMedicine. 2023;94 doi: 10.1016/j.ebiom.2023.104706. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib43] 43.Zheng X., Yao Z., Huang Y., Yu Y., Wang Y., Liu Y., Mao R., Li F., Xiao Y., Wang Y., et al. Deep learning radiomics can predict axillary lymph node status in early-stage breast cancer. Nat. Commun. 2020;11:1236. doi: 10.1038/s41467-020-15027-z. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib44] 44.Liu H., Zou L., Xu N., Shen H., Zhang Y., Wan P., Wen B., Zhang X., He Y., Gui L., Kong W. Deep learning radiomics based prediction of axillary lymph node metastasis in breast cancer. NPJ Breast Cancer. 2024;10:22. doi: 10.1038/s41523-024-00628-4. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib45] 45.Lambin P., Leijenaar R.T.H., Deist T.M., Peerlings J., de Jong E.E.C., van Timmeren J., Sanduleanu S., Larue R.T.H.M., Even A.J.G., Jochems A., et al. Radiomics: the bridge between medical imaging and personalized medicine. Nat. Rev. Clin. Oncol. 2017;14:749–762. doi: 10.1038/nrclinonc.2017.141. [DOI] [PubMed] [Google Scholar]

[bib46] 46.Ma Q., Li Z., Li W., Chen Q., Liu X., Feng W., Lei J. MRI radiomics for the preoperative evaluation of lymphovascular invasion in breast cancer: A meta-analysis. Eur. J. Radiol. 2023;168 doi: 10.1016/j.ejrad.2023.111127. [DOI] [PubMed] [Google Scholar]

[bib47] 47.Dercle L., McGale J., Sun S., Marabelle A., Yeh R., Deutsch E., Mokrane F.Z., Farwell M., Ammari S., Schoder H., et al. Artificial intelligence and radiomics: fundamentals, applications, and challenges in immunotherapy. J. Immunother. Cancer. 2022;10 doi: 10.1136/jitc-2022-005292. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib48] 48.Zhao B. Understanding Sources of Variation to Improve the Reproducibility of Radiomics. Front. Oncol. 2021;11 doi: 10.3389/fonc.2021.633176. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib49] 49.Clarke L.P., Nordstrom R.J., Zhang H., Tandon P., Zhang Y., Redmond G., Farahani K., Kelloff G., Henderson L., Shankar L., et al. The Quantitative Imaging Network: NCI's Historical Perspective and Planned Goals. Transl. Oncol. 2014;7:1–4. doi: 10.1593/tlo.13832. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib50] 50.Kalpathy-Cramer J., Freymann J.B., Kirby J.S., Kinahan P.E., Prior F.W. Quantitative Imaging Network: Data Sharing and Competitive AlgorithmValidation Leveraging The Cancer Imaging Archive. Transl. Oncol. 2014;7:147–152. doi: 10.1593/tlo.13862. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib51] 51.Shukla-Dave A., Obuchowski N.A., Chenevert T.L., Jambawalikar S., Schwartz L.H., Malyarenko D., Huang W., Noworolski S.M., Young R.J., Shiroishi M.S., et al. Quantitative imaging biomarkers alliance (QIBA) recommendations for improved precision of DWI and DCE-MRI derived biomarkers in multicenter oncology trials. J. Magn. Reson. Imag. 2019;49:e101–e121. doi: 10.1002/jmri.26518. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib52] 52.Limkin E.J., Sun R., Dercle L., Zacharaki E.I., Robert C., Reuzé S., Schernberg A., Paragios N., Deutsch E., Ferté C. Promises and challenges for the implementation of computational medical imaging (radiomics) in oncology. Ann. Oncol. 2017;28:1191–1206. doi: 10.1093/annonc/mdx034. [DOI] [PubMed] [Google Scholar]

[bib53] 53.Parmar C., Rios Velazquez E., Leijenaar R., Jermoumi M., Carvalho S., Mak R.H., Mitra S., Shankar B.U., Kikinis R., Haibe-Kains B., et al. Robust Radiomics feature quantification using semiautomatic volumetric segmentation. PLoS One. 2014;9 doi: 10.1371/journal.pone.0102107. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib54] 54.Zhang J., Cui Z., Shi Z., Jiang Y., Zhang Z., Dai X., Yang Z., Gu Y., Zhou L., Han C., et al. A robust and efficient AI assistant for breast tumor segmentation from DCE-MRI via a spatial-temporal framework. Patterns (N Y) 2023;4 doi: 10.1016/j.patter.2023.100826. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib55] 55.Beuque M.P.L., Lobbes M.B.I., van Wijk Y., Widaatalla Y., Primakov S., Majer M., Balleyguier C., Woodruff H.C., Lambin P. Combining Deep Learning and Handcrafted Radiomics for Classification of Suspicious Lesions on Contrast-enhanced Mammograms. Radiology. 2023;307 doi: 10.1148/radiol.221843. [DOI] [PubMed] [Google Scholar]

[bib56] 56.Zhang X., Su G.H., Chen Y., Gu Y.J., You C. Decoding Intratumoral Heterogeneity: Clinical Potential of Habitat Imaging based on Radiomics. Radiology. 2023;309 doi: 10.1148/radiol.232047. [DOI] [PubMed] [Google Scholar]

[bib57] 57.Torre V., Poggio T.A. On edge detection. IEEE Trans. Pattern Anal. Mach. Intell. 1986;8:147–163. doi: 10.1109/tpami.1986.4767769. [DOI] [PubMed] [Google Scholar]

[bib58] 58.Farge M. Wavelet transforms and their applications to turbulence. Annu. Rev. Fluid Mech. 1992;24:395–457. [Google Scholar]

[bib59] 59.Zwanenburg A., Vallières M., Abdalah M.A., Aerts H.J.W.L., Andrearczyk V., Apte A., Ashrafinia S., Bakas S., Beukinga R.J., Boellaard R., et al. The Image Biomarker Standardization Initiative: Standardized Quantitative Radiomics for High-Throughput Image-based Phenotyping. Radiology. 2020;295:328–338. doi: 10.1148/radiol.2020191145. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib60] 60.Kovalev V.A., Kruggel F., Gertz H.J., von Cramon D.Y. Three-dimensional texture analysis of MRI brain datasets. IEEE Trans. Med. Imag. 2001;20:424–433. doi: 10.1109/42.925295. [DOI] [PubMed] [Google Scholar]

[bib61] 61.Zhou J., Luo L.Y., Dou Q., Chen H., Chen C., Li G.J., Jiang Z.F., Heng P.A. Weakly supervised 3D deep learning for breast cancer classification and localization of the lesions in MR images. J. Magn. Reson. Imag. 2019;50:1144–1151. doi: 10.1002/jmri.26721. [DOI] [PubMed] [Google Scholar]

[bib62] 62.Bagherzadeh-Khiabani F., Ramezankhani A., Azizi F., Hadaegh F., Steyerberg E.W., Khalili D. A tutorial on variable selection for clinical prediction models: feature selection methods in data mining could improve the results. J. Clin. Epidemiol. 2016;71:76–85. doi: 10.1016/j.jclinepi.2015.10.002. [DOI] [PubMed] [Google Scholar]

[bib63] 63.Herrero Vicent C., Tudela X., Moreno Ruiz P., Pedralva V., Jiménez Pastor A., Ahicart D., Rubio Novella S., Meneu I., Montes Albuixech Á., Santamaria M.Á., et al. Machine Learning Models and Multiparametric Magnetic Resonance Imaging for the Prediction of Pathologic Response to Neoadjuvant Chemotherapy in Breast Cancer. Cancers. 2022;14 doi: 10.3390/cancers14143508. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib64] 64.Breiman L. Random forests. Mach. Learn. 2001;45:5–32. [Google Scholar]

[bib65] 65.Hearst M.A., Dumais S.T., Osuna E., Platt J., Scholkopf B. Support vector machines. IEEE Intell. Syst. Their Appl. 1998;13:18–28. [Google Scholar]

[bib66] 66.Tibshirani R. Regression shrinkage and selection via the lasso. J. Roy. Stat. Soc. B Stat. Methodol. 1996;58:267–288. [Google Scholar]

[bib67] 67.Kanungo T., Mount D.M., Netanyahu N.S., Piatko C.D., Silverman R., Wu A.Y. An efficient k-means clustering algorithm: Analysis and implementation. IEEE Trans. Pattern Anal. Mach. Intell. 2002;24:881–892. [Google Scholar]

[bib68] 68.Monti S., Tamayo P., Mesirov J., Golub T. Consensus clustering: a resampling-based method for class discovery and visualization of gene expression microarray data. Mach. Learn. 2003;52:91–118. [Google Scholar]

[bib69] 69.Bezdek J.C., Ehrlich R., Full W. FCM: The fuzzy c-means clustering algorithm. Comput. Geosci. 1984;10:191–203. [Google Scholar]

[bib70] 70.Goodfellow I., Pouget-Abadie J., Mirza M., Xu B., Warde-Farley D., Ozair S., Courville A., Bengio Y. Generative adversarial nets. Adv. Neural Inf. Process. Syst. 2014;27 [Google Scholar]

[bib71] 71.Vincent P., Larochelle H., Bengio Y., Manzagol P.-A. 2008. Extracting and Composing Robust Features with Denoising Autoencoders; pp. 1096–1103. [Google Scholar]

[bib72] 72.Joachims T. 1999. Transductive Inference for Text Classification Using Support Vector Machines; pp. 200–209. [Google Scholar]

[bib73] 73.Salimans T., Goodfellow I., Zaremba W., Cheung V., Radford A., Chen X. Improved techniques for training gans. Adv. Neural Inf. Process. Syst. 2016;29 [Google Scholar]

[bib74] 74.Halligan S., Menu Y., Mallett S. Why did European Radiology reject my radiomic biomarker paper? How to correctly evaluate imaging biomarkers in a clinical setting. Eur. Radiol. 2021;31:9361–9368. doi: 10.1007/s00330-021-07971-1. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib75] 75.Kocak B., Durmaz E.S., Erdim C., Ates E., Kaya O.K., Kilickesmez O. Radiomics of Renal Masses: Systematic Review of Reproducibility and Validation Strategies. AJR Am. J. Roentgenol. 2020;214:129–136. doi: 10.2214/ajr.19.21709. [DOI] [PubMed] [Google Scholar]

[bib76] 76.Harrell F.E., Jr., Lee K.L., Califf R.M., Pryor D.B., Rosati R.A. Regression modelling strategies for improved prognostic prediction. Stat. Med. 1984;3:143–152. doi: 10.1002/sim.4780030207. [DOI] [PubMed] [Google Scholar]

[bib77] 77.Heagerty P.J., Lumley T., Pepe M.S. Time-dependent ROC curves for censored survival data and a diagnostic marker. Biometrics. 2000;56:337–344. doi: 10.1111/j.0006-341x.2000.00337.x. [DOI] [PubMed] [Google Scholar]

[bib81] 78.Li H., Mendel K.R., Lan L., Sheth D., Giger M.L. Digital Mammography in Breast Cancer: Additive Value of Radiomics of Breast Parenchyma. Radiology. 2019;291:15–20. doi: 10.1148/radiol.2019181113. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib82] 79.Drukker K., Giger M.L., Joe B.N., Kerlikowske K., Greenwood H., Drukteinis J.S., Niell B., Fan B., Malkov S., Avila J., et al. Combined Benefit of Quantitative Three-Compartment Breast Image Analysis and Mammography Radiomics in the Classification of Breast Masses in a Clinical Data Set. Radiology. 2019;290:621–628. doi: 10.1148/radiol.2018180608. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib83] 80.Bickelhaupt S., Jaeger P.F., Laun F.B., Lederer W., Daniel H., Kuder T.A., Wuesthof L., Paech D., Bonekamp D., Radbruch A., et al. Radiomics Based on Adapted Diffusion Kurtosis Imaging Helps to Clarify Most Mammographic Findings Suspicious for Cancer. Radiology. 2018;287:761–770. doi: 10.1148/radiol.2017170273. [DOI] [PubMed] [Google Scholar]

[bib84] 81.Zhang J., Zhan C., Zhang C., Song Y., Yan X., Guo Y., Ai T., Yang G. Fully automatic classification of breast lesions on multi-parameter MRI using a radiomics model with minimal number of stable, interpretable features. Radiol. Med. 2023;128:160–170. doi: 10.1007/s11547-023-01594-w. [DOI] [PubMed] [Google Scholar]

[bib136] 82.Truhn D., Schrading S., Haarburger C., Schneider H., Merhof D., Kuhl C. Radiomic versus Convolutional Neural Networks Analysis for Classification of Contrast-enhancing Lesions at Multiparametric Breast MRI. Radiology. 2019;290:290–297. doi: 10.1148/radiol.2018181352. [DOI] [PubMed] [Google Scholar]

[bib85] 83.Li H., Zhu Y., Burnside E.S., Huang E., Drukker K., Hoadley K.A., Fan C., Conzen S.D., Zuley M., Net J.M., et al. Quantitative MRI radiomics in the prediction of molecular classifications of breast cancer subtypes in the TCGA/TCIA data set. NPJ Breast Cancer. 2016;2 doi: 10.1038/npjbcancer.2016.12. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib86] 84.Leithner D., Horvat J.V., Marino M.A., Bernard-Davila B., Jochelson M.S., Ochoa-Albiztegui R.E., Martinez D.F., Morris E.A., Thakur S., Pinker K. Radiomic signatures with contrast-enhanced magnetic resonance imaging for the assessment of breast cancer receptor status and molecular subtypes: initial results. Breast Cancer Res. 2019;21:106. doi: 10.1186/s13058-019-1187-z. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib88] 85.Petrillo A., Fusco R., Petrosino T., Vallone P., Granata V., Rubulotta M.R., Pariante P., Raiano N., Scognamiglio G., Fanizzi A., et al. A multicentric study of radiomics and artificial intelligence analysis on contrast-enhanced mammography to identify different histotypes of breast cancer. Radiol. Med. 2024;129:864–878. doi: 10.1007/s11547-024-01817-8. [DOI] [PubMed] [Google Scholar]

[bib137] 86.Ji Y., Whitney H.M., Li H., Liu P., Giger M.L., Zhang X. Differences in Molecular Subtype Reference Standards Impact AI-based Breast Cancer Classification with Dynamic Contrast-enhanced MRI. Radiology. 2023;307 doi: 10.1148/radiol.220984. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib138] 87.Umutlu L., Kirchner J., Bruckmann N.M., Morawitz J., Antoch G., Ingenwerth M., Bittner A.K., Hoffmann O., Haubold J., Grueneisen J., et al. Multiparametric Integrated (18)F-FDG PET/MRI-Based Radiomics for Breast Cancer Phenotyping and Tumor Decoding. Cancers. 2021;13 doi: 10.3390/cancers13122928. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib89] 88.Jiang L., You C., Xiao Y., Wang H., Su G.H., Xia B.Q., Zheng R.C., Zhang D.D., Jiang Y.Z., Gu Y.J., Shao Z.M. Radiogenomic analysis reveals tumor heterogeneity of triple-negative breast cancer. Cell Rep. Med. 2022;3 doi: 10.1016/j.xcrm.2022.100694. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib90] 89.Yu Y., Tan Y., Xie C., Hu Q., Ouyang J., Chen Y., Gu Y., Li A., Lu N., He Z., et al. Development and Validation of a Preoperative Magnetic Resonance Imaging Radiomics-Based Signature to Predict Axillary Lymph Node Metastasis and Disease-Free Survival in Patients With Early-Stage Breast Cancer. JAMA Netw. Open. 2020;3 doi: 10.1001/jamanetworkopen.2020.28086. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib91] 90.Han L., Zhu Y., Liu Z., Yu T., He C., Jiang W., Kan Y., Dong D., Tian J., Luo Y. Radiomic nomogram for prediction of axillary lymph node metastasis in breast cancer. Eur. Radiol. 2019;29:3820–3829. doi: 10.1007/s00330-018-5981-2. [DOI] [PubMed] [Google Scholar]

[bib139] 91.Zhan C., Hu Y., Wang X., Liu H., Xia L., Ai T. Prediction of Axillary Lymph Node Metastasis in Breast Cancer using Intra-peritumoral Textural Transition Analysis based on Dynamic Contrast-enhanced Magnetic Resonance Imaging. Acad. Radiol. 2022;29 doi: 10.1016/j.acra.2021.02.008. S107-s115. [DOI] [PubMed] [Google Scholar]

[bib140] 92.Gao Y., Luo Y., Zhao C., Xiao M., Ma L., Li W., Qin J., Zhu Q., Jiang Y. Nomogram based on radiomics analysis of primary breast cancer ultrasound images: prediction of axillary lymph node tumor burden in patients. Eur. Radiol. 2021;31:928–937. doi: 10.1007/s00330-020-07181-1. [DOI] [PubMed] [Google Scholar]

[bib128] 93.Zhu T., Huang Y.H., Li W., Zhang Y.M., Lin Y.Y., Cheng M.Y., Wu Z.Y., Ye G.L., Lin Y., Wang K. Multifactor artificial intelligence model assists axillary lymph node surgery in breast cancer after neoadjuvant chemotherapy: multicenter retrospective cohort study. Int. J. Surg. 2023;109:3383–3394. doi: 10.1097/js9.0000000000000621. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib92] 94.Qu L., Mei X., Yi Z., Zou Q., Zhou Q., Zhang D., Zhou M., Pei L., Long Q., Meng J., et al. An unsupervised learning model based on CT radiomics features accurately predicts axillary lymph node metastasis in breast cancer patients-diagnostic study. Int. J. Surg. 2024 doi: 10.1097/js9.0000000000001778. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib93] 95.Liu Z., Li Z., Qu J., Zhang R., Zhou X., Li L., Sun K., Tang Z., Jiang H., Li H., et al. Radiomics of Multiparametric MRI for Pretreatment Prediction of Pathologic Complete Response to Neoadjuvant Chemotherapy in Breast Cancer: A Multicenter Study. Clin. Cancer Res. 2019;25:3538–3547. doi: 10.1158/1078-0432.Ccr-18-3190. [DOI] [PubMed] [Google Scholar]

[bib94] 96.Braman N.M., Etesami M., Prasanna P., Dubchuk C., Gilmore H., Tiwari P., Plecha D., Madabhushi A. Intratumoral and peritumoral radiomics for the pretreatment prediction of pathological complete response to neoadjuvant chemotherapy based on breast DCE-MRI. Breast Cancer Res. 2017;19:57. doi: 10.1186/s13058-017-0846-1. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib121] 97.Braman N., Prasanna P., Whitney J., Singh S., Beig N., Etesami M., Bates D.D.B., Gallagher K., Bloch B.N., Vulchi M., et al. Association of Peritumoral Radiomics With Tumor Biology and Pathologic Response to Preoperative Targeted Therapy for HER2 (ERBB2)-Positive Breast Cancer. JAMA Netw. Open. 2019;2 doi: 10.1001/jamanetworkopen.2019.2561. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib95] 98.Antunovic L., De Sanctis R., Cozzi L., Kirienko M., Sagona A., Torrisi R., Tinterri C., Santoro A., Chiti A., Zelic R., Sollini M. PET/CT radiomics in breast cancer: promising tool for prediction of pathological response to neoadjuvant chemotherapy. Eur. J. Nucl. Med. Mol. Imag. 2019;46:1468–1477. doi: 10.1007/s00259-019-04313-8. [DOI] [PubMed] [Google Scholar]

[bib141] 99.Zhang H., Cao W., Liu L., Meng Z., Sun N., Meng Y., Fei J. Noninvasive prediction of node-positive breast cancer response to presurgical neoadjuvant chemotherapy therapy based on machine learning of axillary lymph node ultrasound. J. Transl. Med. 2023;21:337. doi: 10.1186/s12967-023-04201-8. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib142] 100.Mao N., Shi Y., Lian C., Wang Z., Zhang K., Xie H., Zhang H., Chen Q., Cheng G., Xu C., Dai Y. Intratumoral and peritumoral radiomics for preoperative prediction of neoadjuvant chemotherapy effect in breast cancer based on contrast-enhanced spectral mammography. Eur. Radiol. 2022;32:3207–3219. doi: 10.1007/s00330-021-08414-7. [DOI] [PubMed] [Google Scholar]

[bib96] 101.Jiang M., Li C.L., Luo X.M., Chuan Z.R., Lv W.Z., Li X., Cui X.W., Dietrich C.F. Ultrasound-based deep learning radiomics in the assessment of pathological complete response to neoadjuvant chemotherapy in locally advanced breast cancer. Eur. J. Cancer. 2021;147:95–105. doi: 10.1016/j.ejca.2021.01.028. [DOI] [PubMed] [Google Scholar]

[bib97] 102.Huang Y., Zhu T., Zhang X., Li W., Zheng X., Cheng M., Ji F., Zhang L., Yang C., Wu Z., et al. Longitudinal MRI-based fusion novel model predicts pathological complete response in breast cancer treated with neoadjuvant chemotherapy: a multicenter, retrospective study. EClinicalMedicine. 2023;58 doi: 10.1016/j.eclinm.2023.101899. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib98] 103.Fu Y., Lei Y.T., Huang Y.H., Mei F., Wang S., Yan K., Wang Y.H., Ma Y.H., Cui L.G. Longitudinal ultrasound-based AI model predicts axillary lymph node response to neoadjuvant chemotherapy in breast cancer: a multicenter study. Eur. Radiol. 2024 doi: 10.1007/s00330-024-10786-5. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib99] 104.Li W., Huang Y.H., Zhu T., Zhang Y.M., Zheng X.X., Zhang T.F., Lin Y.Y., Wu Z.Y., Liu Z.Y., Lin Y., et al. Noninvasive Artificial Intelligence System for Early Predicting Residual Cancer Burden during Neoadjuvant Chemotherapy in Breast Cancer. Ann. Surg. 2024 doi: 10.1097/sla.0000000000006279. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib143] 105.Park H., Lim Y., Ko E.S., Cho H.H., Lee J.E., Han B.K., Ko E.Y., Choi J.S., Park K.W. Radiomics Signature on Magnetic Resonance Imaging: Association with Disease-Free Survival in Patients with Invasive Breast Cancer. Clin. Cancer Res. 2018;24:4705–4714. doi: 10.1158/1078-0432.Ccr-17-3783. [DOI] [PubMed] [Google Scholar]

[bib100] 106.Wang X., Xie T., Luo J., Zhou Z., Yu X., Guo X. Radiomics predicts the prognosis of patients with locally advanced breast cancer by reflecting the heterogeneity of tumor cells and the tumor microenvironment. Breast Cancer Res. 2022;24:20. doi: 10.1186/s13058-022-01516-0. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib102] 107.Arefan D., Zuley M.L., Berg W.A., Yang L., Sumkin J.H., Wu S. Assessment of Background Parenchymal Enhancement at Dynamic Contrast-enhanced MRI in Predicting Breast Cancer Recurrence Risk. Radiology. 2024;310 doi: 10.1148/radiol.230269. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib78] 108.Geras K.J., Mann R.M., Moy L. Artificial Intelligence for Mammography and Digital Breast Tomosynthesis: Current Concepts and Future Perspectives. Radiology. 2019;293:246–259. doi: 10.1148/radiol.2019182627. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib79] 109.Díaz O., Rodríguez-Ruíz A., Sechopoulos I. Artificial Intelligence for breast cancer detection: Technology, challenges, and prospects. Eur. J. Radiol. 2024;175 doi: 10.1016/j.ejrad.2024.111457. [DOI] [PubMed] [Google Scholar]

[bib80] 110.Mudigonda N.R., Rangayyan R.M., Desautels J.E. Gradient and texture analysis for the classification of mammographic masses. IEEE Trans. Med. Imag. 2000;19:1032–1043. doi: 10.1109/42.887618. [DOI] [PubMed] [Google Scholar]

[bib87] 111.Mazurowski M.A., Zhang J., Grimm L.J., Yoon S.C., Silber J.I. Radiogenomic analysis of breast cancer: luminal B molecular subtype is associated with enhancement dynamics at MR imaging. Radiology. 2014;273:365–372. doi: 10.1148/radiol.14132641. [DOI] [PubMed] [Google Scholar]

[bib101] 112.Li H., Zhu Y., Burnside E.S., Drukker K., Hoadley K.A., Fan C., Conzen S.D., Whitman G.J., Sutton E.J., Net J.M., et al. MR Imaging Radiomics Signatures for Predicting the Risk of Breast Cancer Recurrence as Given by Research Versions of MammaPrint, Oncotype DX, and PAM50 Gene Assays. Radiology. 2016;281:382–391. doi: 10.1148/radiol.2016152110. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib103] 113.Wu J., Mayer A.T., Li R. Integrated imaging and molecular analysis to decipher tumor microenvironment in the era of immunotherapy. Semin. Cancer Biol. 2022;84:310–328. doi: 10.1016/j.semcancer.2020.12.005. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib104] 114.Liu Z., Duan T., Zhang Y., Weng S., Xu H., Ren Y., Zhang Z., Han X. Radiogenomics: a key component of precision cancer medicine. Br. J. Cancer. 2023;129:741–753. doi: 10.1038/s41416-023-02317-8. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib105] 115.Sun K., Zhu H., Chai W., Yan F. Multimodality MRI radiomics analysis of TP53 mutations in triple negative breast cancer. Front. Oncol. 2023;13 doi: 10.3389/fonc.2023.1153261. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib108] 116.Lo Gullo R., Wen H., Reiner J.S., Hoda R., Sevilimedu V., Martinez D.F., Thakur S.B., Jochelson M.S., Gibbs P., Pinker K. Assessing PD-L1 Expression Status Using Radiomic Features from Contrast-Enhanced Breast MRI in Breast Cancer Patients: Initial Results. Cancers. 2021;13 doi: 10.3390/cancers13246273. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib109] 117.Wu Z., Lin Q., Wang H., Chen J., Wang G., Fu G., Li L., Bian T. Intratumoral and Peritumoral Radiomics Based on Preoperative MRI for Evaluation of Programmed Cell Death Ligand-1 Expression in Breast Cancer. J. Magn. Reson. Imag. 2024;60:588–599. doi: 10.1002/jmri.29109. [DOI] [PubMed] [Google Scholar]

[bib106] 118.Su G.H., Jiang L., Xiao Y., Zheng R.C., Wang H., Jiang Y.Z., Peng W.J., Shao Z.M., Gu Y.J., You C. A Multiomics Signature Highlights Alterations Underlying Homologous Recombination Deficiency in Triple-Negative Breast Cancer. Ann. Surg Oncol. 2022;29:7165–7175. doi: 10.1245/s10434-022-11958-7. [DOI] [PubMed] [Google Scholar]

[bib111] 119.Yamamoto S., Maki D.D., Korn R.L., Kuo M.D. Radiogenomic analysis of breast cancer using MRI: a preliminary study to define the landscape. AJR Am. J. Roentgenol. 2012;199:654–663. doi: 10.2214/ajr.11.7824. [DOI] [PubMed] [Google Scholar]

[bib112] 120.Yamamoto S., Han W., Kim Y., Du L., Jamshidi N., Huang D., Kim J.H., Kuo M.D. Breast Cancer: Radiogenomic Biomarker Reveals Associations among Dynamic Contrast-enhanced MR Imaging, Long Noncoding RNA, and Metastasis. Radiology. 2015;275:384–392. doi: 10.1148/radiol.15142698. [DOI] [PubMed] [Google Scholar]

[bib113] 121.Wu J., Li B., Sun X., Cao G., Rubin D.L., Napel S., Ikeda D.M., Kurian A.W., Li R. Heterogeneous Enhancement Patterns of Tumor-adjacent Parenchyma at MR Imaging Are Associated with Dysregulated Signaling Pathways and Poor Survival in Breast Cancer. Radiology. 2017;285:401–413. doi: 10.1148/radiol.2017162823. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib114] 122.Fan M., Xia P., Liu B., Zhang L., Wang Y., Gao X., Li L. Tumour heterogeneity revealed by unsupervised decomposition of dynamic contrast-enhanced magnetic resonance imaging is associated with underlying gene expression patterns and poor survival in breast cancer patients. Breast Cancer Res. 2019;21:112. doi: 10.1186/s13058-019-1199-8. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib115] 123.Liu Q., Hu P. Radiogenomic association of deep MR imaging features with genomic profiles and clinical characteristics in breast cancer. Biomark. Res. 2023;11:9. doi: 10.1186/s40364-023-00455-y. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib107] 124.Shahbandi A., Nguyen H.D., Jackson J.G. TP53 Mutations and Outcomes in Breast Cancer: Reading beyond the Headlines. Trends Cancer. 2020;6:98–110. doi: 10.1016/j.trecan.2020.01.007. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib110] 125.Chopra N., Tovey H., Pearson A., Cutts R., Toms C., Proszek P., Hubank M., Dowsett M., Dodson A., Daley F., et al. Homologous recombination DNA repair deficiency and PARP inhibition activity in primary triple negative breast cancer. Nat. Commun. 2020;11:2662. doi: 10.1038/s41467-020-16142-7. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib116] 126.Li L., Patil D., Petruncio G., Harnden K.K., Somasekharan J.V., Paige M., Wang L.V., Salvador-Morales C. Integration of Multitargeted Polymer-Based Contrast Agents with Photoacoustic Computed Tomography: An Imaging Technique to Visualize Breast Cancer Intratumor Heterogeneity. ACS Nano. 2021;15:2413–2427. doi: 10.1021/acsnano.0c05893. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib117] 127.Su G.H., Xiao Y., You C., Zheng R.C., Zhao S., Sun S.Y., Zhou J.Y., Lin L.Y., Wang H., Shao Z.M., et al. Radiogenomic-based multiomic analysis reveals imaging intratumor heterogeneity phenotypes and therapeutic targets. Sci. Adv. 2023;9 doi: 10.1126/sciadv.adf0837. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib118] 128.Fan M., Wang K., Zhang Y., Ge Y., Lü Z., Li L. Radiogenomic analysis of cellular tumor-stroma heterogeneity as a prognostic predictor in breast cancer. J. Transl. Med. 2023;21:851. doi: 10.1186/s12967-023-04748-6. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib119] 129.Shi Z., Huang X., Cheng Z., Xu Z., Lin H., Liu C., Chen X., Liu C., Liang C., Lu C., et al. MRI-based Quantification of Intratumoral Heterogeneity for Predicting Treatment Response to Neoadjuvant Chemotherapy in Breast Cancer. Radiology. 2023;308 doi: 10.1148/radiol.222830. [DOI] [PubMed] [Google Scholar]

[bib120] 130.Su G.H., Xiao Y., Jiang L., Zheng R.C., Wang H., Chen Y., Gu Y.J., You C., Shao Z.M. Radiomics features for assessing tumor-infiltrating lymphocytes correlate with molecular traits of triple-negative breast cancer. J. Transl. Med. 2022;20:471. doi: 10.1186/s12967-022-03688-x. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib122] 131.Ma J., Chen K., Li S., Zhu L., Yu Y., Li J., Ma J., Ouyang J., Wu Z., Tan Y., et al. MRI-based radiomic models to predict surgical margin status and infer tumor immune microenvironment in breast cancer patients with breast-conserving surgery: a multicenter validation study. Eur. Radiol. 2024;34:1774–1789. doi: 10.1007/s00330-023-10144-x. [DOI] [PubMed] [Google Scholar]

[bib123] 132.Wang Q., Sun K., Liu R., Song Y., Lv Y., Bi P., Yang F., Li S., Zhao J., Li X., et al. Single-cell transcriptome sequencing of B-cell heterogeneity and tertiary lymphoid structure predicts breast cancer prognosis and neoadjuvant therapy efficacy. Clin. Transl. Med. 2023;13 doi: 10.1002/ctm2.1346. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib124] 133.Li K., Ji J., Li S., Yang M., Che Y., Xu Z., Zhang Y., Wang M., Fang Z., Luo L., et al. Analysis of the Correlation and Prognostic Significance of Tertiary Lymphoid Structures in Breast Cancer: A Radiomics-Clinical Integration Approach. J. Magn. Reson. Imag. 2024;59:1206–1217. doi: 10.1002/jmri.28900. [DOI] [PubMed] [Google Scholar]

[bib125] 134.Robinson K., Li H., Lan L., Schacht D., Giger M. Radiomics robustness assessment and classification evaluation: A two-stage method demonstrated on multivendor FFDM. Med. Phys. 2019;46:2145–2156. doi: 10.1002/mp.13455. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib126] 135.Granzier R.W.Y., Verbakel N.M.H., Ibrahim A., van Timmeren J.E., van Nijnatten T.J.A., Leijenaar R.T.H., Lobbes M.B.I., Smidt M.L., Woodruff H.C. MRI-based radiomics in breast cancer: feature robustness with respect to inter-observer segmentation variability. Sci. Rep. 2020;10 doi: 10.1038/s41598-020-70940-z. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib127] 136.Schilsky R.L., Michels D.L., Kearbey A.H., Yu P.P., Hudis C.A. Building a rapid learning health care system for oncology: the regulatory framework of CancerLinQ. J. Clin. Oncol. 2014;32:2373–2379. doi: 10.1200/jco.2014.56.2124. [DOI] [PubMed] [Google Scholar]

[bib129] 137.Friedman J.H. Greedy function approximation: a gradient boosting machine. Ann. Stat. 2001;29:1189–1232. [Google Scholar]

[bib130] 138.Lundberg S.M., Lee S.-I. Advances in neural information processing systems 30. 2017. A unified approach to interpreting model predictions. [Google Scholar]

[bib131] 139.Huang Y., Wang X., Cao Y., Li M., Li L., Chen H., Tang S., Lan X., Jiang F., Zhang J. Multiparametric MRI model to predict molecular subtypes of breast cancer using Shapley additive explanations interpretability analysis. Diagn. Interv. Imaging. 2024;105:191–205. doi: 10.1016/j.diii.2024.01.004. [DOI] [PubMed] [Google Scholar]

[bib132] 140.Selvaraju R.R., Cogswell M., Das A., Vedantam R., Parikh D., Batra D. 2017. Grad-cam: Visual Explanations from Deep Networks via Gradient-Based Localization; pp. 618–626. [Google Scholar]

[bib133] 141.Iqbal S., Qureshi A.N., Alhussein M., Aurangzeb K., Anwar M.S. AD-CAM: Enhancing Interpretability of Convolutional Neural Networks with a Lightweight Framework - From Black Box to Glass Box. IEEE J. Biomed. Health Inform. 2023 doi: 10.1109/jbhi.2023.3329231. [DOI] [PubMed] [Google Scholar]

[bib134] 142.Barnett A.J., Schwartz F.R., Tao C., Chen C., Ren Y., Lo J.Y., Rudin C. A case-based interpretable deep learning model for classification of mass lesions in digital mammography. Nat. Mach. Intell. 2021;3:1061–1070. [Google Scholar]

[bib135] 143.Loizidou K., Elia R., Pitris C. Computer-aided breast cancer detection and classification in mammography: A comprehensive review. Comput. Biol. Med. 2023;153 doi: 10.1016/j.compbiomed.2023.106554. [DOI] [PubMed] [Google Scholar]

PERMALINK

Radiomics in breast cancer: Current advances and future directions

Ying-Jia Qi

Guan-Hua Su

Chao You

Xu Zhang

Yi Xiao

Yi-Zhou Jiang

Zhi-Ming Shao

Summary

Graphical abstract

Introduction

Challenges of the diagnosis and treatment in breast cancer

Epidemiology, diagnosis, and treatment of breast cancer

Current challenges in breast cancer

Radiomics in breast cancer

Imaging modalities

Categories of radiomic features

Workflow of radiomics in breast cancer

Figure 1.

Image acquisition

Tumor segmentation

Feature extraction

Model building and analysis

Utilizing radiomics to predict breast cancer clinicopathological indices and clinical outcomes

Table 1.

Distinguishing benign masses versus malignant tumors

Classification of molecular subtypes and other clinicopathological indices

Prediction of axillary and sentinel lymph node status

Prediction of response to neoadjuvant therapy

Prediction of survival and recurrence

Utilizing radiomics to bridge the gap between phenotypic and microscopic scale information

Table 2.

Predicting molecular events

Uncovering biological alterations

Current challenges of applying radiomics to clinical practice

Generalizability

Interpretability

Convenience

Future directions

Figure 2.

Optimizing validation strategies

Multicenter validation

Prospective clinical trials

Enhancing interpretability

From hand-craft radiomics to DL radiomics

From radiogenomics to radio-multi-omics

Establishing an all-in-one AI-aided radiology-clinic system

Acknowledgments

Author contributions

Declaration of interests

Footnotes

Contributor Information

Supplemental information

References

Associated Data

Supplementary Materials

ACTIONS

PERMALINK

RESOURCES

Similar articles

Cited by other articles

Links to NCBI Databases