Skip to main content
Bentham Open Access logoLink to Bentham Open Access
. 2023 May 31;19(11):1231–1244. doi: 10.2174/1573405619666230123104243

Research on Segmentation Technology in Lung Cancer Radiotherapy Based on Deep Learning

Jun Huang 1, Tao Liu 1, Beibei Qian 1, Zhibo Chen 1, Ya Wang 1,*
PMCID: PMC10351083  PMID: 36694318

Abstract

Background

Lung cancer has the highest mortality rate among cancers. Radiation therapy (RT) is one of the most effective therapies for lung cancer. The correct segmentation of lung tumors (LTs) and organs at risk (OARs) is the cornerstone of successful RT.

Methods

We searched four databases for relevant material published in the last 10 years: Web of Science, PubMed, Science Direct, and Google Scholar. The advancement of deep learning-based segmentation technology for lung cancer radiotherapy (DSLC) research was examined from the perspectives of LTs and OARs.

Results

In this paper, Most of the dice similarity coefficient (DSC) values of LT segmentation in the surveyed literature were above 0.7, whereas the DSC indicators of OAR segmentation were all over 0.8.

Conclusion

The contribution of this review is to summarize DSLC research methods and the issues that DSLC faces are discussed, as well as possible viable solutions. The purpose of this review is to encourage collaboration among experts in lung cancer radiotherapy and DL and to promote more research into the use of DL in lung cancer radiotherapy.

Keywords: Lung cancer, deep learning, image segmentation, organs at risk, lung tumors, radiation therapy

1. INTRODUCTION

1.1. Motivation

Lung cancer is the deadliest cancer in the world [1, 2]. Fig. (1) depicts WHO's global cancer data from 2020, which reveal that there were around 1.8 million fatal cases, the highest mortality rate of all cancer categories [3].

Fig. (1).

Fig. (1)

Global incidence of and deaths from cancer.

In recent years, radiation therapy (RT) has made great technological progress and has played an irreplaceable role in the treatment of lung cancer [4-8]; more than 50% of patients with malignant tumors need to receive RT [9]. The fundamental purpose of RT is to maximize the radiation dose to the target area to kill tumor cells while reducing or avoiding unnecessary radiation to the surrounding organs at risk (OARs). Therefore, the gross tumor volume (GTV), clinical target volume (CTV), and OARs should be accurately segmented in RT planning [10]. At present, automatic segmentation technology based on the atlas is more mature [11-13]; however, the biggest disadvantage of this technology is that it relies heavily on similarities between images. In recent years, several automatic segmentation techniques based on deep learning have been proposed [14-17]. Deep learning (DL) has been widely used in oncology, radiology, and other medical fields to better assist doctors with disease prediction and diagnosis [17-24]. DL in lung cancer radiotherapy segmentation can help doctors not only get more accurate and effective segmentation results [25-31], but also reduce the workload of manually segmenting patient images, allowing them to spend more time on optimizing radiotherapy plans.

1.2. Contribution

In this paper, we investigate the application of DL to radiotherapy in lung cancer, conduct an extensive survey of OAR and lung tumor (LT) segmentation, and compare different segmentation methods based on DL. Section 2 introduces the research strategy of the paper and some commonly used lung cancer datasets and compares this study with related work. Section 3 describes the basic knowledge and evaluation indicators of DL, and focuses on the two clinical application points of LT and OAR segmentation in the process of lung cancer radiotherapy. Section 4 discusses current challenges and possible solutions. Finally, the paper is concluded in Section 5. We investigated many pieces of literature and found that there are few reviews on deep learning-based segmentation technology for lung cancer radiotherapy (DSLC). This paper aims to present the latest developments in DSLC for researchers and provide readers with a convenient reference.

2. LITERATURE SEARCH

A large amount of literature was read during the research for this paper. This section explains the approach and selection criteria for conducting a literature search in detail. There is also a summary of commonly used lung datasets.

2.1. Search Policies and Criteria

We retrieved relevant literature from the last ten years using four databases: Web of Science, PubMed, Science Direct, and Google Scholar. The following keyword combinations were employed in the search process: “Lung cancer,” “Radiotherapy for lung cancer,” “Lung segmentation,” “Lung tumor segmentation,” “Artificial intelligence.” The queried results were imported into Endnote for deduplication [32], and there were 2183 literature items obtained after filtering. In this section, we used Endnote to analyze these papers.

Fig. (2) shows a chart of the relevant literature over the last 10 years. Fig. (2a) shows the publication trend: the number of articles published in this direction increased by 148% in the past three years, but the overall number of articles published remained low. Fig. (2b) is a keyword analysis diagram of papers in related areas in the past 10 years, among which DL accounts for a large proportion of the word cloud. To sum up, the data show that DSLC is a hot research topic that has emerged in recent years.

Fig. (2).

Fig. (2)

Literature research and analysis in the past 10 years. (a) Post trend chart; the x-axis represents the input year and the y-axis represents the number of posts. (b) Keyword analysis chart.

2.2. Literature Survey

Table 1 compares five review articles on OAR and LT segmentation and detection in the past five years. This survey mainly analyzed the deficiencies of the literature in terms of coverage, data indicators, and research trends.

Table 1.

Detailed analysis of our study compared to existing reviews.

Refs. Year Literature Coverage
Range
Type of Learning
Methods
Main Theme Existing
Surveys
are Reviewed
Analyzed Research
Trend
Metrics Details Datasets Details Current Challenge,
Solutions
[33] 2021 2013-2020 Deep Learning Multi-organ segmentation NO YES NO NO NO
[34] 2020 2017-2019 Traditional Multi-organ
tumor detection
NO NO NO NO NO
[35] 2019 2017-2019 Deep Learning Lung cancer
image analysis
NO NO NO NO NO
[36] 2018 2009-2018 Deep Learning
and Traditional
Lung nodule detection NO NO YES NO NO
[37] 2021 2017-2020 Deep Learning Multi-organ segmentation
and lung tumor
segmentation
NO NO NO NO YES
[38] 2022 2013-2021 Deep Learning Lung tumor
segmentation
NO NO NO NO NO
OURS 2022 2018-2021 Deep Learning Multi-organ segmentation
and lung tumor
segmentation
YES YES YES YES YES

The results in Table 1 indicate some limitations in the existing reviews. First, there is a lack of detailed reviews explaining the limitations of other studies and the motivation for their own research; second, there is a lack of analysis of research trends; third, there is a lack of evaluation of relevant research work, metric details, and dataset details; and fourth, there is a lack of discussion of current research challenges and possible solutions. We conducted a detailed DSLC survey in an attempt to fill the gaps in the existing literature.

2.3. Common Datasets for Lung Tumors

Some publicly available datasets are frequently used in the diagnosis and treatment of lung cancer using deep neural networks, as shown in (Table 2).

Table 2.

Public lung tumor datasets.

Dataset Year Input Details Refs.
Non-small
cell lung cancer (NSCLC)
2021 CT/ PET-CT 285,411 images with a total data capacity of 97.6 GB [39]
LIDC-IDRI 2020 CT/DX/CR 244,527 images with a total data capacity of 125 GB [40]
Lung CT Segmentation Challenge 2017 2020 CT 9593 images with a total data capacity of 4.8 GB [41]
NIH 2019 CT 32,735 images with a total data capacity of 221 GB [42]
NLST 2017 CT Over 75,000 CT images in 15 sub-databases [43]
Data Science Bowl 2017 2017 CT The National Cancer Institute’s Center for Cancer Research provides a two-stage dataset; the data capacity of the first stage exceeds 66 GB, and that of the second stage exceeds 38 GB [44]
ChestX-ray14 2017 X-Ray 112,120 images with a total data capacity of 45 GB [45]
QIN LUNG CT 2017 CT 3954 images with a total data capacity of 2.08 GB [46]
LUNA16 2016 CT 888 CT images of 1084 tumors [47]
SPIE-AAPM Lung CT Challenge 2016 CT 22,489 images with a total data capacity of 12.1 GB [48]
LungCT-Diagnosis 2014 CT 4682 images with a total data capacity of 2.5 GB [49]
TCIA 2021 MRI/CT Large-scale public database containing medical images such as common tumors and corresponding clinical information, with all data organized and managed by TCIA. [50]
TCGA 2021 MRI/CT 11,961 lung cases with a total data capacity of 2.5 PB [51]
CLEF 2017 2017 N/A CLEF dataset includes 500 patients, categorized into five TB types: invasive, focal, tubercular, miliary, and cavernous fibroma. [52]
SCR 2000 X-Ray 247 chest X-rays, including left and right lung, left and right clavicle, heart, etc. Total data capacity of 2.5 MB [53]
JSRT 2000 CT/X-Ray 154 conventional CT chest radiographs, the total data capacity of 1.33 GB [54]

Among the lung cancer datasets listed in Table 2, the LIDC-IDRI dataset provides an authoritative and open standard for research on lung nodules [55, 56], and the details of other lung cancer-related datasets are also summarized in the table.

3. DEEP LEARNING AUTOMATIC SEGMENTATION TECHNOLOGY

3.1. Introduction to Deep Learning

DL has been widely used in image analysis in pathology [57-59]. The current popular DL algorithm includes a convolutional neural network (CNN) [60] and generative adversarial network (GAN) [61]; the latter has the characteristics of unsupervised learning [62]. Some scholars integrate GAN and CNN for medical image segmentation. Based on the wider application of CNN, in this paper, we focus on the application of CNN in DSLC. CNN contains convolutional, pooling, and fully connected layers. The role of the convolutional layer is to use the convolution kernel to extract features from the input image. The role of the pooling layer is to reduce the resolution of the feature map and the consumption of memory. The role of the fully connected layer is to classify and output the extracted features. The structure diagram of CNN is shown in Fig. (3).

Fig. (3).

Fig. (3)

Convolutional neural network (CNN) structure diagram.

Commonly used basic CNNs are VGG [63] and ResNet [64]. VGG is a network model with a simple structure and strong generalization ability. VGG increases the receptive field by stacking small convolution kernels. ResNet is based on the concept of using shortcut connections to solve the problem of deep network degradation so that thousands of layers of convolutional networks can converge.

In addition to the basic convolutional network, there are two commonly used segmentation neural networks, FCN [65] and U-Net [66]. FCN uses a skip connection structure to fuse the shallow appearance information and deep semantic information of the feature map to segment images more accurately. U-Net has a better processing effect for medical image data with a small amount of data, large image size, blurred boundaries, and multi-modal imagery, and has become the baseline for most medical image semantic segmentation tasks. In addition, the derived Attention U-Net [67] further improves the performance of image segmentation.

3.2. Common Evaluation Indicators

Table 3 lists the metrics commonly used in experiments; among them, the dice similarity coefficient (DSC) is a simple and useful statistical validation metric that can be applied to study the accuracy of image segmentation [68].

Table 3.

Evaluation parameters.

graphic file with name CMIM-19-1231_T3.jpg

3.3. LT and OAR Segmentation for Lung Cancer

Patients with advanced lung cancer have a five-year survival rate of less than 15%, but survival rates after treatment for early-stage lung cancer can range from 40 to 70% [75]. As a result, early detection and treatment are critical to increasing the cure rate [76]. The primary treatment method for lung cancer is RT. In clinical practice, precise irradiation of tumor target areas and protection of OARs are critical factors for RT success, and DSLC plays an important role in these tasks. This section discusses and compares DSLC-related work from two perspectives: LT segmentation and OAR segmentation (Fig. 4).

Fig. (4).

Fig. (4)

Segmentation of lung tumors (LTs) and organs at risk (OARs) for lung cancer.

3.3.1. Lung Tumor Segmentation

In the diagnosis of clinical lung tumors (LTs), it is often necessary to process images of different modalities, such as X-ray, computed tomography (CT), ultrasound, magnetic resonance imaging (MRI), positron emission tomography (PET), and positron emission computed tomography (PET-CT), as shown in Fig. (4).

Zhang et al. [77] developed an improved ResNet for segmenting of non-small-cell lung tumors on CT images, combining shallow and deep semantic features to produce dense pixel output. In 2020, Pang et al. [78] proposed CTumorGAN, a unified end-to-end adversarial learning framework, for the prediction of CT images using multi-level supervision of different modules to deal with problems such as class imbalance, small tumors, and label noise, with a DSC coefficient of 71.08%. With a success rate of 99.92%, the method improves the model's generalization ability for different objective functions and achieves a stable tumor segmentation scheme with a low error rate. Jiang J. et al. [79] developed a cross-modal (MR-CT) depth learning segmentation method, which enhances training data by converting manually segmented CT images into pseudo-MR images.

MRI provides high resolution for soft tissue, allowing a better view of tumors and adjacent normal tissues. Wang et al. [29] presented A-Net, a new patient-specific adaptive convolutional neural network that uses MRI imags and GTV annotation to train the network model; its DSC index and precision are 0.82 0.10 and 0.81 0.08, respectively Jiang et al. [80] developed a cross-modality induced distillation method for cone-beam CT (CBCT) images. The idea is to use MRI to guide the training of the CBCT segmentation network.

The advantage of PET is that it can accurately locate small tumors and distinguish benign and malignant tumors early. Leung et al. [81] proposed mU-Net for segmenting of PET images, which is designed to help address the challenge of a lack of clinical training data with known ground-truth tumor boundaries in PET.

PET-CT combines the high sensitivity of PET images with the anatomical information of CT images and overcomes the difficulties of blurred image boundaries, low contrast, and complex backgrounds. Zhao et al. [74] proposed a multimodal segmentation method based on 3D full convolution neural network, which can extract the characteristic information of PET and CT simultaneously for tumor segmentation, and has strong robustness. In 2020, Li et al. [82] integrated CT tumor probability maps and PET images into a recognition model, which could accurately identify the input images. In 2021, Lei et al. [83] proposed a recurrent fusion network (RFN) for automatic PET-CT tumor segmentation that can complementarily fuse the intermediate segmentation results to obtain multi-modal image features, which improves the convergence speed. Fu et al. [84] proposed a multi-modal spatial attention network module (MSAM).

In addition, Bi et al. [85] established a deep expansion residual network based on ResNet-101, which is used to automatically sketch the CTV of lung cancer patients undergoing radiotherapy after surgery. The experimental results show that, compared with manual contour, the effect of deep learning assisted sketching is better, and 35% of the time is saved than before. Jemaa et al. [86] proposed an end-to-end method to quickly identify and segment tumors by combining 2D and 3D convolutional networks, which can adapt to an extreme imbalance between healthy tissue volumes and heterogeneity of input images. Jiang et al. [87] developed two multiresolution residual connection networks, combined the features and functional levels of multiple image resolutions, and detected and segmented lung tumors through residual connection. After evaluation, it can accurately segment the volume of lung tumors.

Table 4 lists the lung tumor segmentation work in detail. Fig. (5) shows the DSC accuracy of lung tumor segmentation in the related literature, where the abscissa represents the reference numbers in Table 4 and the ordinate represents the DSC values, which are mostly above 0.7 [88].

Table 4.

Selected works on deep learning-based automated segmentation of lung tumors (LTs).

Team DataSets Input Net Evaluation Metrics Research Highlights
Wang et al. (2018) [29] 9 patients MRI ANet DSC PRE SN
0.82 ± 0.10 0.82± 0.08 0.85 ± 0.13
Adaptive neural network, A-Net, was introduced to delineate LTs
Zhao et al. (2018) [74] 84 patients PET-CT 3D FCN DSC 0.85 ± 0.08 Novel multimodal segmentation network, 3D FCN, proposed to integrate PET and CT images into the same utility
Jiang et al. (2018) [87] 1210 patients
TCIA
MSKCC
LIDC
CT MRRN DSC HD95% SN PRE
TCIA 0.74 7.94 0.80 0.73
MSKC 0.75 5.85 0.82 0.72
LIDC 0.68 2.60 0.85 0.67
Multi-resolution residual connection network proposed to combine features across multiple image resolution through residual connections to detect and segment LTs
Bi et al. (2019) [85] 269 patients CT ResNet DSC CV SDD
0.75 ± 0.06 0.129±0.04 0.47±0.22
ResNet network used to segment LTs, obtaining better results than manual segmentation
Jiang J.et al. (2019) [79] 28 patients MR -CT U-Net +
cross modality
DSC VR HD95%
0.75 ±0.12 0.19±0.15 9.36±6.00
Cross-modal deep learning (DL) segmentation method used to better segment LTs
Li et al. (2020) [82] 84 patients PET-CT 3D FCN DSC VE CE
0.86 ± 0.05 0.16±0.12 0.30±0.12
CT tumor probability maps and PET intensity images combined for accurate multimodal tumor segmentation
Zhang et al. (2020) [77] 330 patients CT ResNet DSC JS SN
0.73±0.07 0.68±0.09 0.74±0.07
Fast segmentation of LTs using improved ResNet
Pang et al. (2020) [78] NSCLC CT CTumor-GAN DSC PRE SN
0.7108 0.7734 0.7042
Authors propose CTumorGAN algorithm for better segmentation of LTs
Leung et al. (2020) [81] 30 patients PET mU-Net DSC JS HD95%
0.73 0.65 ±0.02 3.25±0.30
mU-Net algorithm was used to segment smaller LTs on slices of PET
Jemaa et al. (2020) [86] 3664 trial
scans
PET-CT 2D and 3D architecture Lymphoma Lung
SN 0.926 0.93
DSC 0.886 N/A
Combined with 2D and 3D convolutional networks, rapid detection and segmentation of LTs were realized
Fu et al. (2021) [84] 876 NSCLC
3063 soft tissue sarcoma (STS) [88]
PET-CT MSAM NSCLC STS
DSC 0.7144 0.6226
PRE 0.7293 0.69
SN 0.8109 0.6494
SP 0.9995 0.9974
Based on attention mechanism, a multimodal spatial attention network module (MSAM) is proposed to strengthen learning of tumor-related
Lei et al. (2021) [83] 70 patients PET-CT RFN DSC 0.6775±0.2341
PRE 0.7164±0.2728
SN 0.7318±0.2558
RFN network is proposed to segment LTs
Fig. (5).

Fig. (5)

Segmentation dice similarity coefficient (DSC) for lung tumors.

3.3.2. Organ-at-Risk Segmentation

Because RT can affect organs outside the target area, radiation oncologists must accurately segment OARs to reduce the probability of normal tissue complications after RT. DL segmentation models can now automatically segment OARs based on trial and error. This section discusses various methods for solving the difficult problem of automatic OAR segmentation, such as experimenting with different network architectures, introducing loss functions, and combining supervised and unsupervised learning methods, which will be discussed in detail below. Zhu et al. [89] improved the deep learning split network based on U-Net, which can split many kinds of OARs in the lung. Among them, the DSC index for segmenting the lung is the highest, reaching 95%. Feng et al. [73] proposed a based 3D U-Net model to automatically segment five sternal OARs, including the left and right lungs, heart, esophagus, and spinal cord. Based on U-Net, Vesal et al. [90] used the expansion convolution and aggregation residual connection methods to segment OARs in chest CT images, and achieved high-precision segmentation of 20 undiscovered test samples.

GAN [61] can produce quite good output through mutual game learning of generative and discriminative models. Dong et al. [91] proposed a UNet-GAN strategy to automatically delineate the left and right lungs, spinal cord, esophagus, and heart. With the assistance of adversarial networks, the segmentation accuracy was greatly improved. It has been found in experiments that the traditional convolutional neural network model is not very compatible with medical imaging. He et al. [92] proposed a unified encoder–decoder architecture based on the U-Net model and used it in multi-task procedures. It is trained in learning mode, and the experimental results show that the DSC accuracy on the heart reaches 95%.

Zhao et al. [64] introduced multi-instance loss and conditional adversarial loss based on the FCN network to solve the segmentation problem under more severe pathological conditions, and the experiment obtained a DSC of 97.93%. Chen et al. [93] designed a weighted DSC based on the loss function of the coefficients is used to solve the problem of segmentation imbalance, and the experiment obtained a DSC of 97.55%.

The biggest challenge of DL in the medical field is the lack of annotated training sets. Hu et al. [94] used the Mask R-CNN architecture to combine supervised and unsupervised machine learning methods to automatically segment lungs on CT images and obtained the best results for lung segmentation. Research on automatic segmentation of OARs is not only important for radiotherapy but also provides inspiration and implications for other image segmentation algorithms.

Harten et al. [95] proposed various segmentation technologies based on different frameworks in combination with 2D-CNN and 3D-CNN to automatically segment four OARs: heart, aorta, trachea, and esophagus. The experimental results show that the best performance is achieved in DSC and HD. Akila et al. [96] proposed a convolutional deep wide network (CDWN) to segment lung regions in thoracic CT images. In the experiment, the DSC and ACC of the LIDC-IDRI dataset reached 95% and 98% respectively. Zhang, et al. [97] established a CNN network based on ResNet-101 for automatic segmentation of OARs, including lungs, esophagus, heart, liver, and spinal cord.

Table 5 details related work on OAR segmentation. Fig. (6) depicts the DSC accuracy for OAR segmentation in the searched literature, where the abscissa represents the reference numbers in Table 5 and the ordinate represents the DSC values, which are mostly above 0.8.

Table 5.

Selected works on deep learning-based automated segmentation of OARs for lung cancer.

Team DataSets In OARs Network DSC Metric Other Evaluations Metric Research
Highlights
Zhao et al.
(2018) [64]
LIDC-IDRI
CLEF
HUG [98]
CT Lung FCN LIDC:0.9176
CLEF:0.9613
HUG:0.9793
N/A Introduced multi-instance loss and conditional adversarial loss functions
Zhu et al.
(2019) [89]
66 case of CT
(30 case train,
36 case test)
CT Lung
Heart
Esophagus
Spinal cord
liver
AdaptedU-Net Lung:0.95±0.01
Esophagus:0.71±0.05
Spinal cord:0.79±0.03
Heart:0.91±0.03
Liver:0.89±0.02
Lung:
1.93±0.51(MSD) 7.96±2.57(HD 95%)
Esophagus:
2.18±0.80(MSD) 7.83±2.85(HD 95%)
Spinal cord:
1.25±0.23(MSD) 4.01±2.05(HD 95%)
Heart:
2.92±1.51(MSD) 7.98±4.56(HD 95%)
Liver:3.21±0.93( MSD)
Encoder–decoder U-Net neural network constructed based on convolutional neural networks (CNN) to automatically segment OARs
Harten et al.
(2019) [95]
Seg Thor (60 thoracic CT scans) CT Heart
Aorta
Trachea
esophagus
CNN Esophagus:0.84±0.05
Heart:0.94±0.02
Aorta:0.93±0.01
Trachea:0.91±0.02
HD
Esophagus 3.4±2.3
Heart 2.0±1.1
Aorta 2.7±3.6
Trachea 2.1±1.0
2DCNN and 3DCNN frameworks combined to segment multiple organs at risk on chest CT
Akila et al.
(2020) [96]
LIDC-IDRI CT Lung CDWN 0.95 ± 0.03 JS:0.91 ± 0.04 ACC0.98 ± 0.01
SP:0.99±0.01
SN:0.95 ±0.03 PRE:0.95±0.03
CDWN proposed for segmentation of lung regions in chest CT images
Dong et al.
(2019) [91]
35 patiens CT Left lung
Right lung
Heart
Esophagus
Spinal cord
Unet-GAN Left Lung0.97±0.01
Right Lung0.97±0.01
Esophagus0.75±0.08
Spinal cord0.90±0.04
Heart0.87±0.05
Left Lung:
0.61±0.73(MSD) 0.9989±0.0010(SP)
0.97±0.02(SN) 2.07±1.93(HD 95%)
Right Lung:
0.65±0.53(MSD) 0.9992±0.0007(SP)
0.96±0.02(SN) 2.50±3.34(HD 95%)
Esophagus:
1.05±0.66(MSD) 4.52±3.81(HD 95%)
Spinal cord:
0.38±0.27(MSD) 1.19±0.46(HD 95%)
Heart:
1.49±0.85( MSD) 4.58±3.67(HD 95%)
U-Net used as generator and FCN as discriminator to design U-Net generative adversarial network (U-Net-GAN) to segment OARs in lung CT images
Feng et al.
(2019) [73]
60 thoracic CT scans CT Left lung
Right lung
Heart
Esophagus
Spinal cord
3D U-
Net
Left Lung0.98±0.01
Right Lung0.97±0.02
Esophagus0.73±0.09
Spinal cord0.89±0.04
Heart0.93±0.02
Left Lung:
0.59±0.29(MSD) 2.10±0.94(HD 95%)
Right Lung:
0.93±0.57(MSD) 3.96±2.85(HD 95%)
Esophagus:
2.34±2.38(MSD) 8.71±10.59(HD 95%)
Spinal cord:
0.66±0.25(MSD) 1.89±0.63(HD 95%)
Heart:
2.30±0.49(MSD) 6.57±1.50(HD 95%)
Novel DCNN method proposed for automatic segmentation of chest OARs from large 3D images
Hu
et al.
(2020) [94]
1265
images
CT Lung Mask
R-CNN
0.9733 ± 0.0324 SN:0.97 ± 0.09
SP:0.9711 ± 0.0365
Improved lung segmentation performance using a combination of Mask R-CNN and K-means
Vesal et al.
(2019) [90]
60 patients CT Heart
Aorta
Trachea
esophagus
2D Unet Esophagus:0.858
Heart 0.941
Aorta 0.938
Trachea0.926
HD
Esophagus 0.331
Heart 0.226
Aorta 0.297
Trachea 0.193
Introduced extended convolution in two-dimensional U-Net to better segment OARs
He et al.
(2019) [92]
SegTHOR
[99]
CT Heart
Aorta
Trachea
esophagus
Unet Esophagus:0.8594
Heart:0.9500
Aorta: 0.9484
Trachea:0.9201
HD
Esophagus 0.2743
Heart 0.1383
Aorta 0.1129
Trachea 0.1824
Optimized false positive filtering algorithm to decrease number of falsely segmented organ pixels
Zhang et al.
(2020) [97]
250 patients CT Left lung
Right lung
Heart
Esophagus
Spinal cord
liver
AS-CNN Left lung0.948±0.013
Left lung0.943±0.015
Esophagus0.732±0.069
Spinal cord0.821±0.046
Heart0.893±0.048
liver0.937±0.027
MSD:
Left Lung1.10±0.15
Right Lung2.23±2.33
Esophagus1.38±0.44
Spinal cord0.87±0.21
Heart 1.65±0.48
Liver 2.03±1.49
AS-CNN algorithm proposed, proving that DL is better than atlas method in automatic organ segmentation
Chen et al.
(2019) [93]
45 thorax DECT DECT Left lung
Right lung
Liver
Spleen
Left Kidney
Right Kidney
3D FCN L_lung0.975±0.0064
R_lung0.976±0.0161
Liver0.962±0.0164
Spleen:0.914±0.0486
L_Kidney:0.937±0.0312
R_Kidney:0.945±0.0122
HD
Left Lung: 6.97±2.67
Right Lung: 8.08±3.51
Liver: 9.64±4.89
Spleen:6.93±3.44
Left Kidney:4.41±2.17
Right Kidney:3.62±1.75
Multiple 3D CNNs proposed for segmentation of multi-organ DECT images
Fig. (6).

Fig. (6)

Segmentation DSC for OARs.

4. DISCUSSION

Although recent studies show that DSLC outperforms traditional segmentation methods in terms of efficiency and accuracy [100], it still faces some challenges.

4.1. Medical Imaging Problems

Tissues and organs in medical images have a high degree of similarity, especially in low-contrast images, where the segmentation target is very similar to the background and it is difficult to distinguish the boundaries. In terms of medicine, MRI images are preferable to CT as input because they provide better visualization [101]. In computer technology, new algorithms can be developed for solving the low-contrast problem of medical image segmentation. For example, 3D algorithmic networks should be used because they can adequately extract contextual spatial information from medical images compared to 2D networks, alleviating the problem of low contrast [102].

4.2. Dataset Size Issue

Obtaining medical images involves patient privacy issues, and the production of medical datasets requires professional doctors to label them. These two reasons lead to a scarcity of large medical datasets. However, training the model without a large number of samples hurts the robustness of the DL algorithm, resulting in overfitting of the trained model, and the small dataset cannot demonstrate the algorithm's generalization ability. These issues make the clinical application of DSLC more difficult. Moreover, apart from the datasets provided by some competitions with common standards, the datasets used by most researchers are of uneven quality, and the datasets created using specific scenarios to verify the overall performance of the algorithms are not convincing. In particular, most DSLC studies are based on single-point dataset training, which lacks diversity, and medical images in real situations have great differences due to race, age, gender, disease, etc., resulting in decreased model segmentation accuracy.

In light of the scarcity of medical datasets, various medical institutions could build large-scale datasets by sharing data in order to provide DL researchers with more expert annotated data under the premise of protecting patient privacy [103]. From the perspective of computer technology, DL researchers can also try to use transfer learning strategies [104] to pre-train network models as a way to alleviate the problem of limited data. Furthermore, medical image datasets can also be augmented by cropping, rotating, filling, and color-enhancing images through data augmentation methods.

4.3. Algorithmic Model Problems

The deeper the layers of the network model, the stronger the ability to extract features and the more complex the network structure. For the pixel-by-pixel classification task of lung images, expanding the number of layers of the network model is conducive to training a more accurate segmentation model. In addition, in order to extract and fuse multi-scale features of images, most researchers try to use more strategies for extracting features in the network, which undoubtedly increases the complexity of the network structure. As the number of network layers of the model increases, the ability to extract features, the data occupied by the GPU memory, and the time to train the model increase at the same time. Most algorithms reduce the training time by sacrificing a large amount of GPU space. This is not a long-term solution, and complex network structure has become a technical barrier limiting the improvement of model segmentation accuracy. It is worth considering how to strike a balance between network design, computing time, and cost. Hu et al. [94] used the improved Mask R-CNN architecture to achieve high-precision segmentation in DSC and combined it with the K-means method to improve the segmentation accuracy while reducing the model structure. At the same time, to avoid the constraints of GPU memory, we can try to use algorithms such as GAN to generate training data artificially to reduce the number of hidden layers or parameters of the network and to overcome hardware constraints to a certain extent.

4.4. Clinical Application Issues

The biggest difference between clinical medical applications and the experimental process is that there will be various unpredictable clinical situations [105]. If the DSLC only operates in a data environment similar to the training dataset, it will be difficult to respond correctly to emergencies. DSLC is required to be able to continuously learn to cope with clinical emergencies. In addition, DL algorithms also lack interpretability, it is difficult to fully understand which factors in the algorithm will lead to degraded segmentation performance, and it may not be possible to control the stability of OAR segmentation and GTV accuracy. If this uncertainty is used in clinical practice, it is very dangerous. Before DSLC is used clinically, relevant hospital personnel should conduct a thorough risk assessment, consider legal and ethical responsibilities, think about measures to deal with emergencies, and formulate a set of detailed standard procedures to protect the safety of patients. Computer-related researchers can also explore new network frameworks that enable models to learn experiences autonomously under unknown conditions, improve models' continuous learning ability, and reduce clinical application risks.

CONCLUSION

In this paper, we investigated many kinds of studies, extracted common datasets and evaluation indicators for LTs, reviewed the basic theory of DL-related algorithms, and discussed and compared DSLC-related work from two aspects of LT and OAR segmentation. By improving the network framework and the segmentation accuracy, DSLC achieved satisfactory results in OAR segmentation of the lung and heart. However, it also has some challenges. To address these challenges, this paper presents an analysis and possible solutions. The author's knowledge is limited, and some important works may not be included in this paper. Hopefully, this review will deepen researchers’ understanding of lung cancer RT and DL, and stimulate collaboration between the two communities to develop a more specialized adjuvant lung cancer RT application system.

ACKNOWLEDGEMENTS

Y.W. conceptualized and presented the idea; J.H., T.L. studied the state-of-the-art methods and summarized the literature and prepared the draft; All authors provided feedback, helped shape the research, and contributed to the final manuscript. All authors have read and agreed to the published version of the manuscript.

LIST OF ABBREVIATIONS

ACC

Accuracy

CBCT

Cone-beam CT

CE

Classification Error

CNN

Convolutional Neural Networks

CT

Computed Tomography

CTV

Clinical Target Volume

DL

Deep Learning

DSC

Dice Similarity Coefficient

DSLC

Deep Learning-Based Segmentation Technology for Lung Cancer Radiotherapy

GAN

Generative Adversarial Network

GTV

Gross Tumor Volume

HD95%

Hausdorff_Distance95%

JS

Jaccard Similarity

LTs

Lung Tumors

MRI

Magnetic Resonance Imaging

MSAM

Multi-Modal Spatial Attention Network Module

MSD

The Mean Surface Distance

NSCLC

Non-Small Cell Lung Cancer

OARs

Organs at Risk

PET

Positron Emission Tomography

CONSENT FOR PUBLICATION

Not applicable.

FUNDING

This work was financially supported by the Natural Science Foundation of Anhui Provincial (Grant No. 1808085MF202), the Talent project of Anhui Provincial (Grant No. gxgwfx200050), the Natural Science Research Project of Anhui Provincial (Grant No. KJ2021A0662), the Science Research and Innovation Team of Fuyang Normal University (Grant No. kytd202004), the Natural Science Research Project of Fuyang Normal University (Grant No. 2018kyqd0028, 2021FSKJ02ZD). My project name is AnHui Provincial Graduate Innovation and Entrepreneurship Practice Project, and Grant No. 2022cxcysj189.

CONFLICT OF INTEREST

The authors have no conflicts of interest, financial or otherwise.

REFERENCES

  • 1.Bray F., Laversanne M., Weiderpass E., Soerjomataram I. The ever increasing importance of cancer as a leading cause of premature death worldwide. Cancer. 2021;127(16):3029–3030. doi: 10.1002/cncr.33587. [DOI] [PubMed] [Google Scholar]
  • 2.Sung H., Ferlay J., Siegel R.L., et al. Global cancer statistics 2020: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J. Clin. 2021;71(3):209–249. doi: 10.3322/caac.21660. [DOI] [PubMed] [Google Scholar]
  • 3.Latest global cancer data: Cancer burden rises to 19.3 million new cases and 10.0 million cancer deaths in 2020. 2020. Available from: https://web.archive.org/web/202 20301120144/ https://www.iarc.who.int/wp-content/upl oads/2020/12/pr292_E.pdf(Accessedon: 12 february 2022).
  • 4.Vinod S.K., Hau E. Radiotherapy treatment for lung cancer: Current status and future directions. Respirology. 2020;25(Suppl. 2):61–71. doi: 10.1111/resp.13870. [DOI] [PubMed] [Google Scholar]
  • 5.Nagata Y., Kimura T. Stereotactic body radiotherapy (SBRT) for Stage I lung cancer. Jpn. J. Clin. Oncol. 2018;48(5):405–409. doi: 10.1093/jjco/hyy034. [DOI] [PubMed] [Google Scholar]
  • 6.Brown S., Banfill K., Aznar M.C., Whitehurst P., Faivre Finn C. The evolving role of radiotherapy in non-small cell lung cancer. Br. J. Radiol. 2019;92(1104):20190524. doi: 10.1259/bjr.20190524. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 7.Burdett S., Rydzewska L., Tierney J., et al. Postoperative radiotherapy for non small cell lung cancer. Cochrane Database System Rev. 2016;10(10):CD002142. doi: 10.1002/14651858.CD002142.pub3. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 8.Baker S., Dahele M., Lagerwaard F.J., Senan S. A critical review of recent developments in radiotherapy for non-small cell lung cancer. Radiat. Oncol. 2016;11(1):115. doi: 10.1186/s13014-016-0693-8. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 9.Meyer P., Noblet V., Mazzara C., Lallement A. Survey on deep learning for radiotherapy. Comput. Biol. Med. 2018;98:126–146. doi: 10.1016/j.compbiomed.2018.05.018. [DOI] [PubMed] [Google Scholar]
  • 10.Samarasinghe G., Jameson M., Vinod S., et al. Deep learning for segmentation in radiation therapy planning: a review. J. Med. Imaging Radiat. Oncol. 2021;65(5):578–595. doi: 10.1111/1754-9485.13286. [DOI] [PubMed] [Google Scholar]
  • 11.Daisne J.F., Blumhofer A. Atlas-based automatic segmentation of head and neck organs at risk and nodal target volumes: a clinical validation. Radiat. Oncol. 2013;8(1):154. doi: 10.1186/1748-717X-8-154. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 12.Cabezas M., Oliver A., Lladó X., Freixenet J., Bach Cuadra M. A review of atlas-based segmentation for magnetic resonance brain images. Comput. Methods Programs Biomed. 2011;104(3):e158–e177. doi: 10.1016/j.cmpb.2011.07.015. [DOI] [PubMed] [Google Scholar]
  • 13.Bai W., Shi W., Ledig C., Rueckert D. Multi-atlas segmentation with augmented features for cardiac MR images. Med. Image Anal. 2015;19(1):98–109. doi: 10.1016/j.media.2014.09.005. [DOI] [PubMed] [Google Scholar]
  • 14.Meiburger K.M., Acharya U.R., Molinari F. Automated localization and segmentation techniques for B-mode ultrasound images: A review. Comput. Biol. Med. 2018;92:210–235. doi: 10.1016/j.compbiomed.2017.11.018. [DOI] [PubMed] [Google Scholar]
  • 15.Wang Y, Zhao L, Wang M, Song Z. Organ at risk segmentation in head and neck ct images using a two-stage segmentation framework based on 3D U-Net. IEEE Access . 2019;7:144591–602. [Google Scholar]
  • 16.Liu C., Gardner S.J., Wen N., et al. Automatic segmentation of the prostate on CT images using deep neural networks (DNN). Int. J. Radiat. Oncol. Biol. Phys. 2019;104(4):924–932. doi: 10.1016/j.ijrobp.2019.03.017. [DOI] [PubMed] [Google Scholar]
  • 17.Men K., Zhang T., Chen X., et al. Fully automatic and robust segmentation of the clinical target volume for radiotherapy of breast cancer using big data and deep learning. Phys. Med. 2018;50:13–19. doi: 10.1016/j.ejmp.2018.05.006. [DOI] [PubMed] [Google Scholar]
  • 18.Avanzo M., Stancanello J., Pirrone G., Sartor G. Radiomics and deep learning in lung cancer. Strahlenther. Onkol. 2020;196(10):879–887. doi: 10.1007/s00066-020-01625-9. [DOI] [PubMed] [Google Scholar]
  • 19.Liu Z., Yao C., Yu H., Wu T. Deep reinforcement learning with its application for lung cancer detection in medical Internet of Things. Future Gener. Comput. Syst. 2019;97:1–9. doi: 10.1016/j.future.2019.02.068. [DOI] [Google Scholar]
  • 20.Polat H., Danaei Mehr H. Classification of pulmonary CT images by using hybrid 3D-deep convolutional neural network architecture. Appl. Sci. (Basel) 2019;9(5):940. doi: 10.3390/app9050940. [DOI] [Google Scholar]
  • 21.Vrtovec T., Močnik D., Strojan P., Pernuš F., Ibragimov B. Auto segmentation of organs at risk for head and neck radiotherapy planning: From atlas based to deep learning methods. Med. Phys. 2020;47(9):e929–e950. doi: 10.1002/mp.14320. [DOI] [PubMed] [Google Scholar]
  • 22.Kholiavchenko M., Sirazitdinov I., Kubrak K., et al. Contour-aware multi-label chest X-ray organ segmentation. Int. J. CARS. 2020;15(3):425–436. doi: 10.1007/s11548-019-02115-9. [DOI] [PubMed] [Google Scholar]
  • 23.Tamang L.D., Kim B.W. Deep learning approaches to colorectal cancer diagnosis: A review. Appl. Sci. (Basel) 2021;11(22):10982. doi: 10.3390/app112210982. [DOI] [Google Scholar]
  • 24.Cao H., Liu H., Song E., et al. A two-stage convolutional neural networks for lung nodule detection. IEEE J. Biomed. Health Inform. 2020;24(7):1. doi: 10.1109/JBHI.2019.2963720. [DOI] [PubMed] [Google Scholar]
  • 25.Wong J., Fong A., McVicar N., et al. Comparing deep learning-based auto-segmentation of organs at risk and clinical target volumes to expert inter-observer variability in radiotherapy planning. Radiother. Oncol. 2020;144:152–158. doi: 10.1016/j.radonc.2019.10.019. [DOI] [PubMed] [Google Scholar]
  • 26.Men K., Dai J., Li Y. Automatic segmentation of the clinical target volume and organs at risk in the planning CT for rectal cancer using deep dilated convolutional neural networks. Med. Phys. 2017;44(12):6377–6389. doi: 10.1002/mp.12602. [DOI] [PubMed] [Google Scholar]
  • 27.Souza J.C., Bandeira Diniz J.O., Ferreira J.L., França da Silva G.L., Corrêa Silva A., de Paiva A.C. An automatic method for lung segmentation and reconstruction in chest X-ray using deep neural networks. Comput. Methods Programs Biomed. 2019;177:285–296. doi: 10.1016/j.cmpb.2019.06.005. [DOI] [PubMed] [Google Scholar]
  • 28.Shaziya H., Shyamala K., Zaheer R. Automatic lung segmentation on thoracic CT scans using U-net convolutional network.2018 . International conference on communication and signal processing (ICCSP). 3-5 April 2018; Chennai, India: IEEE; :0643–7. doi: 10.1109/ICCSP.2018.8524484. [DOI] [Google Scholar]
  • 29.Wang C., Tyagi N., Rimner A., et al. Segmenting lung tumors on longitudinal imaging studies via a patient-specific adaptive convolutional neural network. Radiother. Oncol. 2019;131:101–107. doi: 10.1016/j.radonc.2018.10.037. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 30.Han M, Yao G, Zhang W, et al. Segmentation of CT thoracic organs by multi-resolution VB-nets. >SegTHOR@ ISBI; 8-11 April Venice, Italy. 2019 [Google Scholar]
  • 31.Park J., Yun J., Kim N., et al. Fully automated lung lobe segmentation in volumetric chest CT with 3D U-Net: validation with intra-and extra-datasets. J. Digit. Imaging. 2020;33(1):221–230. doi: 10.1007/s10278-019-00223-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 32.Hupe M. EndNote X9. J. Electron. Resour. Med. Libr. 2019;16(3-4):117–119. doi: 10.1080/15424065.2019.1691963. [DOI] [Google Scholar]
  • 33.Fu Y., Lei Y., Wang T., Curran W.J., Liu T., Yang X. A review of deep learning based methods for medical image multi-organ segmentation. Phys. Med. 2021;85:107–122. doi: 10.1016/j.ejmp.2021.05.003. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 34.Sharif M.I., Li J.P., Naz J., Rashid I. A comprehensive review on multi-organs tumor detection based on machine learning. Pattern Recognit. Lett. 2020;131:30–37. doi: 10.1016/j.patrec.2019.12.006. [DOI] [Google Scholar]
  • 35.Wang S., Yang D.M., Rong R., et al. Artificial intelligence in lung cancer pathology image analysis. Cancers (Basel) 2019;11(11):1673. doi: 10.3390/cancers11111673. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 36.Zhang G., Jiang S., Yang Z., et al. Automatic nodule detection for lung cancer in CT images: A review. Comput. Biol. Med. 2018;103:287–300. doi: 10.1016/j.compbiomed.2018.10.033. [DOI] [PubMed] [Google Scholar]
  • 37.Liu X., Li K.W., Yang R., Geng L.S. Review of deep learning based automatic segmentation for lung cancer radiotherapy. Front. Oncol. 2021;2021:11717039. doi: 10.3389/fonc.2021.717039. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 38.Kao Y.S., Yang J. Deep learning-based auto-segmentation of lung tumor PET/CT scans: a systematic review. Clin. Transl. Imaging. 2022;10(2):217–223. doi: 10.1007/s40336-022-00482-z. [DOI] [Google Scholar]
  • 39.Smith K., Nolan T. NSCLC Radiogenomics. Available from: https://web.archive.org/web/20220301 091241/ https://wiki.cancerimagingarchive.net/display/Public/NSCLC+Radiogenomics (Accessed on: 11 February 2022).
  • 40.Vendt B., Nolan T. The Lung Image Database Consortium image collection. Available from: https://wiki.cancerimagingarchive.net/display/Public/LIDC-IDRI (Accessed on: 3 January 2022).
  • 41.Nolan T., Jarosz Q. Lung CT segmentation challenge. 2017. Available from: ive.org/web/20220301131137/ https://wiki.cancerimagingarchive.net/display/Public/Lung+CT+Segmentation+Challenge+2017(Accessed on: 18 February 2022).
  • 42.DeepLesion. Available from: ive.org/web/20220301131812/ https://nihcc.app.box.com/v/DeepLesion (Accessed on: 9 February 2022).
  • 43.NLST Datasets. Available from: rchive.org/web/20220301140337/ https://cdas.cancer.gov/datasets/nlst/ (Accessed on: 5 February 2022).
  • 44.Data Science Bowl. 2017. Available from: https://www.kaggle.com/c/data-science-bowl-2017/(Accessed on: 30 December 2021).
  • 45.NIH Chest X-rays. 2022. Available from: https://www.kaggle.com/nih-chest-xrays/data/ (Accessed on: 12 February 2022).
  • 46.Smith K., Nolan T., QIN Lung CT. Available from: https://wiki.cancerimagingarchive.net/display/Public/QIN+LUNG+CT/ (Accessed on: 15 february 2022).
  • 47.Lung Nodule Analysis. 2016. Available from: https://web.archive.org/web/20220301142254/ https://luna16.grand-challenge.org/Data/(Accessed on: 8 February 2022).
  • 48.Kirby J., Jarosz Q. SPIE-AAPM Lung CT Challenge. Available from: https://web.archive.org/w eb/20220301142410/ https://wiki.cancerimagingarchive.net/display/Public/SPIE-AAPM+Lung+CT+Challenge(Accessed on: 12 January 2022).
  • 49.Clark K., Jarosz Q. LungCT-Diagnosis. Available from: https://web.archive.org/web/202 2030 1143334/ https://wiki.cancerimagingarchive.net/display/Public/LungCT-Diagnosis (Accessed on: 13 February 2022).
  • 50.Web Archive. The cancer imaging archive. Available from: https://web.archive.org/web/20220302020242/ https://www.cancerimagingarchive.net/(Accessed on: 13 February 2022).
  • 51.TCGA. The Cancer Genome Atlas Program. Available from: https://web.archive.org/web/2022 0301143553/ https://www.cancer.gov/about-nci/organiz ation/ccg/research/structural-genomics/tcga (Accessed on: 11 February 2022).
  • 52.ImageCLEF/LifeCLEF-Multimedia Retrieval in CLEF. Available from: https://web.archive.org/web/20220301143736/ https://www.imageclef.org/2017/tuberculosis(Accessed on: 3 January 2022).
  • 53. SCR database: Segmentation in chest radiographs. Available from: https://web.archive.org/web/20220302073253/ https://www.isi.uu.nl/Research/Databases/SCR/ (Accessed on: 10 January 2022).
  • 54.JSRT. Database. Available from: rchive.org/web/20220302073541/ http://db.jsrt.or.jp/eng.php (Accessed on: 10 January 2022).
  • 55.Armato S.G., III, McLennan G., Bidaut L., et al. The Lung Image Database Consortium (LIDC) and Image Database Resource Initiative (IDRI): A completed reference database of lung nodules on CT scans. Med. Phys. 2011;38(2):915–931. doi: 10.1118/1.3528204. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 56.Setio A.A.A., Traverso A., de Bel T., et al. Validation, comparison, and combination of algorithms for automatic detection of pulmonary nodules in computed tomography images: The LUNA16 challenge. Med. Image Anal. 2017;42:1–13. doi: 10.1016/j.media.2017.06.015. [DOI] [PubMed] [Google Scholar]
  • 57.Sahiner B., Pezeshk A., Hadjiiski L.M., et al. Deep learning in medical imaging and radiation therapy. Med. Phys. 2019;46(1):e1–e36. doi: 10.1002/mp.13264. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 58.Zhou X, Li C, Rahaman MM, et al. A comprehensive review for breast histopathology image analysis using classical and deep neural networks.. IEEE Access . 2020;8:90931–56. doi: 10.1109/ACCESS.2020.2993788. [DOI] [Google Scholar]
  • 59.Lee L.K., Liew S.C., Thong W.J. A review of image segmentation methodologies in medical image. Adv Comput Commun Eng Technol. 2015;315(1069):80. doi: 10.1007/978-3-319-07674-4_99. [DOI] [Google Scholar]
  • 60.LeCun Y., Kavukcuoglu K., Farabet C. Convolutional networks and applications in vision.; Proceedings of 2010 IEEE international symposium on circuits and systems; 30 May-2 June 2010; Paris, France: IEEE. pp. 253–6. [Google Scholar]
  • 61.Goodfellow I., Pouget-Abadie J., Mirza M., et al. Generative adversarial nets. Adv. Neural Inf. Process. Syst. 2014;2014:27. [Google Scholar]
  • 62.Raza K., Singh N.K. A tour of unsupervised deep learning for medical image analysis. Curr. Med. Imaging Rev. 2021;17(9):1059–1077. doi: 10.2174/1573405617666210127154257. [DOI] [PubMed] [Google Scholar]
  • 63.Simonyan K, Zisserman A. Very deep convolutional networks forlarge-scale image recognition. arXiv 2014;2014 14091556. [Google Scholar]
  • 64.Zhao T., Gao D., Wang J., Yin Z. 2018 IEEE 15th International Symposium on Biomedical Imaging (ISBI 2018). 24 May 2018; Washington DC, USA: IEEE; Lung segmentation in CT images using a fully convolutional neural network with multi-instance and conditional adversary loss. pp. 505–9. [DOI] [Google Scholar]
  • 65.Long J., Shelhamer E., Darrell T. Fully convolutional networks for semantic segmentation.; Proceedings of the IEEE conference on computer vision and pattern recognition; Boston, MA, USA. 2015. pp. 3431–40. [Google Scholar]
  • 66.Ronneberger O., Fischer P., Brox T. U-Net: Convolutional networks for biomedical image segmentation. In: Navab N,, Hornegger j., Wells W., Frangi A., editors. Medical Image Computing and Computer-Assisted Intervention – MICCAI 2015: MICCAI 2015. Lecture Notes in Computer Science, Vol.9351. Springer, Cham. [DOI] [Google Scholar]
  • 67.Liu T., Qian B., Wang Y., Xie Q. U-Net medical image segmentation based on attention mechanism combination. Int Conf Cogn Inform Proc Appl (CIPA) 2021;2021:805–813. doi: 10.1007/978-981-16-5857-0_103. [DOI] [Google Scholar]
  • 68.Zou K.H., Warfield S.K., Bharatha A., et al. Statistical validation of image segmentation quality based on a spatial overlap index1. Acad. Radiol. 2004;11(2):178–189. doi: 10.1016/S1076-6332(03)00671-8. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 69.Anwar S.M., Majid M., Qayyum A., Awais M., Alnowami M., Khan M.K. Medical image analysis using convolutional neural networks: a review. J. Med. Syst. 2018;42(11):226. doi: 10.1007/s10916-018-1088-1. [DOI] [PubMed] [Google Scholar]
  • 70.Havaei M., Davy A., Warde-Farley D., et al. Brain tumor segmentation with deep neural networks. Med. Image Anal. 2017;35:18–31. doi: 10.1016/j.media.2016.05.004. [DOI] [PubMed] [Google Scholar]
  • 71.Yuan Y., Chao M., Lo Y.C. Automatic skin lesion segmentation using deep fully convolutional networks with jaccard distance. IEEE Trans. Med. Imaging. 2017;36(9):1876–1886. doi: 10.1109/TMI.2017.2695227. [DOI] [PubMed] [Google Scholar]
  • 72.Kumar Y., Gupta S., Singla R., Hu Y-C. A systematic review of artificial intelligence techniques in cancer prediction and diagnosis. Arch. Comput. Methods Eng. 2021;2021:1–28. doi: 10.1007/s11831-021-09648-w. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 73.Feng X., Qing K., Tustison N.J., Meyer C.H., Chen Q. Deep convolutional neural network for segmentation of thoracic organs‐at‐risk using cropped 3D images. Med. Phys. 2019;46(5):2169–2180. doi: 10.1002/mp.13466. [DOI] [PubMed] [Google Scholar]
  • 74.Zhao X., Li L., Lu W., Tan S. Tumor co-segmentation in PET/CT using multi-modality fully convolutional neural network. Phys. Med. Biol. 2018;64(1):015011. doi: 10.1088/1361-6560/aaf44b. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 75.Siegel R.L., Miller K.D., Fuchs H.E., Jemal A. Cancer statistics, 2022. CA Cancer J. Clin. 2022;72(1):7–33. doi: 10.3322/caac.21708. [DOI] [PubMed] [Google Scholar]
  • 76.Sheng K. Artificial intelligence in radiotherapy: a technological review. Front. Med. 2020;14(4):431–449. doi: 10.1007/s11684-020-0761-1. [DOI] [PubMed] [Google Scholar]
  • 77.Zhang F., Wang Q., Li H. Automatic segmentation of the gross target volume in non-small cell lung cancer using a modified version of resNet. Technol. Cancer Res. Treat. 2020;19:1533033820947484. doi: 10.1177/1533033820947484. [DOI] [Google Scholar]
  • 78.Pang S., Du A., Orgun M.A., et al. CTumorGAN: a unified framework for automatic computed tomography tumor segmentation. Eur. J. Nucl. Med. Mol. Imaging. 2020;47(10):2248–2268. doi: 10.1007/s00259-020-04781-3. [DOI] [PubMed] [Google Scholar]
  • 79.Jiang J., Hu Y.C., Tyagi N., et al. Cross modality (CTMRI) prior augmented deep learning for robust lung tumor segmentation from small MR datasets. Med. Phys. 2019;46(10):4392–4404. doi: 10.1002/mp.13695. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 80.Jiang J., Riyahi Alam S., Chen I., et al. Deep cross modality (MR CT) educed distillation learning for cone beam CT lung tumor segmentation. Med. Phys. 2021;48(7):3702–3713. doi: 10.1002/mp.14902. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 81.Leung K.H., Marashdeh W., Wray R., et al. A physics-guided modular deep-learning based automated framework for tumor segmentation in PET. Phys. Med. Biol. 2020;65(24):245032. doi: 10.1088/1361-6560/ab8535. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 82.Li L., Zhao X., Lu W., Tan S. Deep learning for variational multimodality tumor segmentation in PET/CT. Neurocomputing. 2020;392:277–295. doi: 10.1016/j.neucom.2018.10.099. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 83.Bi L., Fulham M., Li N., et al. Recurrent feature fusion learning for multi-modality pet-ct tumor segmentation. Comput. Methods Programs Biomed. 2021;2021:203106043. doi: 10.1016/j.cmpb.2021.106043. [DOI] [PubMed] [Google Scholar]
  • 84.Fu X., Bi L., Kumar A., Fulham M., Kim J. Multimodal spatial attention module for targeting multimodal PET-CT lung tumor segmentation. IEEE J. Biomed. Health Inform. 2021;25(9):3507–3516. doi: 10.1109/JBHI.2021.3059453. [DOI] [PubMed] [Google Scholar]
  • 85.Bi N., Wang J., Zhang T., et al. Deep learning improved clinical target volume contouring quality and efficiency for postoperative radiation therapy in non-small cell lung cancer. Front. Oncol. 2019;9:1192. doi: 10.3389/fonc.2019.01192. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 86.Jemaa S., Fredrickson J., Carano R.A.D., Nielsen T., de Crespigny A., Bengtsson T. Tumor segmentation and feature extraction from whole-body FDG-PET/CT using cascaded 2D and 3D convolutional neural networks. J. Digit. Imaging. 2020;33(4):888–894. doi: 10.1007/s10278-020-00341-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 87.Jiang J., Hu Y.C., Liu C.J., et al. Multiple resolution residually connected feature streams for automatic lung tumor segmentation from CT images. IEEE Trans. Med. Imaging. 2019;38(1):134–144. doi: 10.1109/TMI.2018.2857800. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 88.Vallières M., Freeman C.R., Skamene S.R., El Naqa I. A radiomics model from joint FDG-PET and MRI texture features for the prediction of lung metastases in soft-tissue sarcomas of the extremities. Phys. Med. Biol. 2015;60(14):5471–5496. doi: 10.1088/0031-9155/60/14/5471. [DOI] [PubMed] [Google Scholar]
  • 89.Zhu J., Zhang J., Qiu B., Liu Y., Liu X., Chen L. Comparison of the automatic segmentation of multiple organs at risk in CT images of lung cancer between deep convolutional neural network-based and atlas-based techniques. Acta Oncol. 2019;58(2):257–264. doi: 10.1080/0284186X.2018.1529421. [DOI] [PubMed] [Google Scholar]
  • 90.Vesal S, Ravikumar N, Maier A. A 2D dilated residual U-Net for multi-organ segmentation in thoracic CT. arXiv. 2019;2019:190507710. [Google Scholar]
  • 91.Dong X., Lei Y., Wang T., et al. Automatic multiorgan segmentation in thorax CT images using U‐net‐GAN. Med. Phys. 2019;46(5):2157–2168. doi: 10.1002/mp.13458. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 92.He T., Hu J., Song Y., Guo J., Yi Z. Multi-task learning for the segmentation of organs at risk with label dependence. Med. Image Anal. 2020:61101666. doi: 10.1016/j.media.2020.101666. [DOI] [PubMed] [Google Scholar]
  • 93.Chen S., Zhong X., Hu S., et al. Automatic multiorgan segmentation in dual energy CT (DECT) with dedicated 3D fully convolutional DECT networks. Med. Phys. 2020;47(2):552–562. doi: 10.1002/mp.13950. [DOI] [PubMed] [Google Scholar]
  • 94.Hu Q, de F, Souza LF, Holanda GB, et al. An effective approach for CT lung segmentation using mask region-based convolutional neural networks. Artif. Intell. Med. 2020;2020:103101792. doi: 10.1016/j.artmed.2020.101792. [DOI] [PubMed] [Google Scholar]
  • 95.van Harten LD, Noothout JM, Verhoeff JJ, Wolterink JM, Isgum I. Automatic segmentation of organs at risk in thoracic CT scans by combining 2D and 3D convolutional neural networks.SegTHOR@ ISBI 2019;2019 139099960. [Google Scholar]
  • 96.Akila Agnes S., Anitha J., Dinesh Peter J. Automatic lung segmentation in low-dose chest CT scans using convolutional deep and wide network (CDWN). Neural Comput. Appl. 2020;32(20):15845–15855. doi: 10.1007/s00521-018-3877-3. [DOI] [Google Scholar]
  • 97.Zhang T., Yang Y., Wang J., et al. Comparison between atlas and convolutional neural network based automatic segmentation of multiple organs at risk in non-small cell lung cancer. Medicine (Baltimore) 2020;99(34):e21800. doi: 10.1097/MD.0000000000021800. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 98.Cid YD, Del Toro OAJ, Depeursinge A, Müller H. VISCERAL Challenge@ ISBI 16-19 April 2015; NY, USA. Efficient and fully automatic segmentation of the lungs in CT volumes. pp. 31–5. [Google Scholar]
  • 99.Lambert Z., Petitjean C., Dubray B., Kuan S. Tenth International Conference on Image Processing Theory, Tools and Applications (IPTA). 9-12 November ; Paris, France: IEEE. 2020. pp. 1–6. [Google Scholar]
  • 100.Choi R.Y., Coyner A.S., Kalpathy-Cramer J., Chiang M.F., Campbell J.P. Introduction to machine learning, neural networks, and deep learning. Transl. Vis. Sci. Technol. 2020;9(2):14–4. doi: 10.1167/tvst.9.2.14. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 101.Winfield J.M., Payne G.S., deSouza N.M. Functional MRI and CT biomarkers in oncology. Eur. J. Nucl. Med. Mol. Imaging. 2015;42(4):562–578. doi: 10.1007/s00259-014-2979-0. [DOI] [PubMed] [Google Scholar]
  • 102.Fechter T., Adebahr S., Baltas D., Ben Ayed I., Desrosiers C., Dolz J. Esophagus segmentation in CT via 3D fully convolutional neural network and random walk. Med. Phys. 2017;44(12):6341–6352. doi: 10.1002/mp.12593. [DOI] [PubMed] [Google Scholar]
  • 103.Giovannini S., Macchi C., Liperoti R., et al. Association of body fat with health-related quality of life and depression in nonagenarians: The mugello study. J. Am. Med. Dir. Assoc. 2019;20(5):564–568. doi: 10.1016/j.jamda.2019.01.128. [DOI] [PubMed] [Google Scholar]
  • 104.Lin X., Jiao H., Pang Z., et al. Lung cancer and granuloma identification using a deep learning model to extract 3-dimensional radiomics features in CT imaging. Clin. Lung Cancer. 2021;22(5):e756–e766. doi: 10.1016/j.cllc.2021.02.004. [DOI] [PubMed] [Google Scholar]
  • 105.Coraci D., Giovannini S., Loreti C., Fusco A., Padua L. Management of neuropathic pain: A graph theory based presentation of literature review. Breast J. 2020;26(3):581–582. doi: 10.1111/tbj.13622. [DOI] [PubMed] [Google Scholar]

Articles from Current Medical Imaging are provided here courtesy of Bentham Science Publishers

RESOURCES