Predicting immunohistochemical expression status and molecular subtypes using synthetic magnetic resonance imaging and multiple multiplexed sensitivity encoding diffusion weighted imaging in breast cancer: a prospective study

Abstract

To evaluate and compare the performance of synthetic magnetic resonance imaging (SyMRI) and multiple multiplexed sensitivity encoding diffusion weighted imaging (MUSE DWI) in predicting the expression status of immunohistochemistry (IHC) markers and molecular subtypes in patients with breast cancer. This prospective study included patients who underwent breast SyMRI examinations between September 2022 and April 2024, with subsequent pathological confirmation of breast cancer. SyMRI imaging provided pre-contrast and post-contrast T1, T2, and proton density (PD) relaxation times for breast lesions. We subsequently computed the differences and ratios of these relaxation times before and after contrast enhancement, enabling a comparison of the quantitative values. Univariate analysis was then conducted to examine the associations between each parameter and receptor status, proliferation rate, as well as the subtypes of breast cancer, including Luminal, triple-negative breast cancer (TNBC), and human epidermal growth factor receptor 2 (HER2)-enriched categories. We employed multivariate logistic regression analysis for the purposes of feature selection and model development, subsequently constructing a nomogram based on the optimal model. The multivariate analysis revealed that T2-Pre was the sole quantitative parameter associated with TNBC and luminal breast cancer, whereas the apparent diffusion coefficient (ADC) emerged as an independent predictor for the HER2-enriched subtype. When distinguishing molecular subtypes, area under the curve (AUC) of the joint model which combination of quantitative parameter and other parameter showed a relatively higher AUC than any single model. There was a significant difference in AUCs between joint parameters (T2-Pre + histologic grade) and T2-Pre for distinguishing luminal subtype from non-luminal breast cancers, AUC being 0.878 and 0.709, respectively (P = 0.002). No statistically significant difference was observed in AUC between the paired models for distinguishing HER2-enriched subtypes (joint model combining ADC values and histologic grade versus ADC values alone, AUC = 0.824 and AUC = 0.760, respectively; P > 0.05) and TNBC subtypes (joint model combining T2-weighted pre-contrast imaging (T2-Pre) and enhancement versus T2-Pre alone, AUC = 0.854 and AUC = 0.814, respectively; P > 0.05). Quantitative parameters derived from SyMRI and MUSE DWI represent a promising approach for differentiating breast cancer subtypes. However, these findings are preliminary due to sample size limitations and lack of external validation, requiring confirmation in multicenter studies.

Introduction

Breast cancer is the most prevalent malignancy among women. Identification of prognostic factors is the cornerstone for treating breast cancer¹. Estrogen receptor (ER) and progesterone receptor (PR) are present in normal breast tissue, where they regulate breast development and function through their corresponding nuclear receptors. The continued expression of ER and PR in breast cancer patients indicates that the tumor remains hormonally regulated, and hence can be treated with endocrine therapy². High expression of Ki-67 is indicative of increased cellular proliferation and significantly correlates with the development, metastasis, and prognosis of breast cancer. Human epidermal growth factor receptor 2 (HER2) is involved in regulating cell proliferation and differentiation³ patients with HER2-positive status face a heightened risk of recurrence and metastasis, and may benefit from anti-HER2 targeted therapy. Immunohistochemistry (IHC) can reflect the heterogeneity within the tumor. The expression of IHC including ER, PR, HER2 and Ki-67 is an important basis for molecular typing of breast cancer⁴. Different molecular subtypes of breast cancer have differences in treatment methods, sensitivity of treatment response and prognosis evaluation. Triple-negative breast cancer (TNBC) is typically characterized by the absence of expression of ER, PR, and HER2, showing high sensitivity to chemotherapy. Given the invasiveness of obtaining receptor status and its potential changes during treatment, a repeatable non‑invasive method is urgently needed.

Magnetic resonance imaging (MRI) is a crucial diagnostic tool for breast cancer due to its high soft tissue contrast and cross-sectional display of lesion information⁵. Additionally, dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) can reveal the morphological features and vascular perfusion of breast lesions. However, traditional MRI lack objective and quantitative criteria, and the scanning time is long.

Diffusion-weighted imaging (DWI) is a technique that measures the mean rate of self-diffusion of water molecules. The quantification of DWI is achieved through the apparent diffusion coefficient (ADC). Compared with the single shot echo plane (SS-EPI) DWI, the DWI sequence based on multiple sensitivity encoding (MUSE) achieves high-resolution diffusion imaging through multiple excitation acquisition, which improves the signal-to-noise ratio of the image and improves the reliability of ADC diagnosis^6,7. For imaging of the brain and abdomen, MUSE improved geometric fidelity and image quality in comparison to SS-EPI, so that it has better lesion detectability and improves clinical results^8–10. However, previous studies on ADC differences across breast cancer molecular subtypes have yielded inconsistent results^11–14.

The emerging synthetic magnetic resonance imaging (SyMRI) can obtain multiple T1, T2 and PD contrast-weighted images through one scan, and quantify the intrinsic relaxation time and proton density of the tissue¹⁵. SyMRI‑derived T1 and T2 values show no significant difference from those obtained by conventional multi‑echo spin‑echo sequences¹⁶. SyMRI has been applied to the study of head and neck, bone and joint, and pelvic system diseases^17–20. In the field of breast imaging, studies have demonstrated that quantitative assessment values obtained from SyMRI play a key role in distinguishing between benign and malignant breast lesions^21–24.

Therefore, this study aimed to investigate the value of SyMRI and MUSE DWI in predicting IHC markers and molecular subtypes of breast cancer to provide a reference for clinical treatment.

Materials and methods

Patient population

The study was a non-randomized, prospective, single-center research project approved by our Institutional Review Board. Written informed consent was obtained from all participants. All methods were performed in accordance with the relevant guidelines and regulations. The inclusion criteria were as follows: (1) pathologically confirmed breast cancer, and complete data of ER, PR, HER-2 and Ki-67 indexes detected by IHC; (2) patients with no surgery or neoadjuvant treatment before MRI examination; (3) complete data of SyMRI that allowed for quantitative measurement. Exclusion criteria: (1) previous puncture biopsy or surgical procedures conducted before the MRI examination; (2) lesion diameter < 10 mm; (3) documented history of breast cancer or other malignant tumors before hospital admission.

Image acquisition

All breast MR images were taken using a 3.0T MRI scanner (SIGNA Architect, GE Healthcare, USA). Prior to scanning, the patient lay in a prone position, feet first, with both breasts naturally lowered in the 8-channel phased array breast coil. The scan sequences included routine T1-weighted imaging, fat-suppressed T2-weighted imaging, MUSE DWI (b = 800s/mm²), SS-EPI DWI (b = 800s/mm²), differential subsampling with cartesian ordering (DISCO), and MAGnetic resonance image Compilation (MAGiC). MAGiC is a synthetic MRI technique based on a multiple-dynamic multiple-echo sequence that acquires multiple contrast images and quantifies T1, T2, and proton density (PD) from a single scan. MAGiC was performed before and after contrast enhancement with identical parameters.

Image analysis

Two radiologists (with 6 and 11 years of experience) independently analyzed images and performed quantitative measurements blinded to pathological findings.

Clinical and pathological data collection.

The study involved the collection of essential clinical and pathological information from patients, including age, histologic grade, histologic type, and presence of axillary lymph node metastasis. Additionally, following the American College of Radiology BI-RADS MRI 2015 edition guidelines²⁵, conventional MRI images were assessed and analyzed for their size, shape (round or oval, lobulated, and irregular), margin (spiculated or non-spiculated), T2 signal intensity (low-iso and high), enhancement(rim or non-rim), and time-signal intensity curve (TIC) pattern (persistent, plateau and washout). Divergent opinions were resolved by consensus.

MUSE DWI analysis.

MUSE DWI were imported into GE workstation, and ADC maps were automatically generated by READY View post-processing platform. The region of interest (ROI) was outlined along the edges of the largest level of the lesion with reference to the T2WI and enhanced DISCO image, and the ADCs were measured by avoiding the areas of necrosis and haemorrhage that were visible to the naked eye. The average values were obtained after measuring the ADCs three times each physician.

SyMRI analysis.

The SyMRI were processed by ITK-SNAP software (Version 3.8, ITK-SNAP, http://www.itksnap.org). The original MAGiC images were post-processed, and T1, T2 and PD maps were automatically generated. With reference to the location and extent of the lesion on the dynamically enhanced MRI, the entire lesion was manually outlined as the ROI on the MAGiC images before and after enhancement, taking care to avoid non-tumor areas such as glands. The software automatically reproduced the same ROIs on the pseudo-colour maps of the T1, T2 and PD images before and after enhancement, and generated the parameter values. The parameters before enhancement were denoted as T1-Pre, T2-Pre, PD-Pre, after enhancement were denoted as T1-Gd, T2-Gd, PD-Gd, and the mean value was obtained by outlining each lesion three times by each reader. Subsequently, ∆T1 (T1-Pre – T1-Gd), ∆T2 (T2-Pre – T2-Gd), ∆PD (PD-Pre –PD-Gd), T1-ratio (T1-Pre – T1-Gd/T1-Pre), T2-ratio (T2-Pre – T2-Gd/T2-Pre), and PD-ratio (PD-Pre – PD-Gd/PD-Pre) were manually calculated.

Observer reliability assessment.

Intra‑observer reliability was assessed by the same radiologist repeating ROI segmentation after one month. Inter-observer reliability was assessed by the second radiologist performing independent segmentation. The intraclass correlation coefficient (ICC) was calculated for all parameters. ICC > 0.75 indicated good consistency (0.75–0.9 = good, > 0.9 = excellent).

Histopathological results

All lesions were surgically excised and the IHC status of each surgical specimen were evaluated. According to the ASCO/USCAP guidelines and the Allred score^26,27, IHC was employed to assess the status of ER, PR, HER-2, and Ki-67. Tumors with HER2 status of 3 + or HER2 gene amplification confirmed by fluorescence in situ hybridization were considered HER2 positive. Ki-67 high expression was defined as nuclear staining ≥ 14%, and low expression as < 14%⁴. Based on the status of ER, PR, HER2, and Ki-67 expression level, breast cancer was classified into 5 molecular subtypes: Luminal A (i.e. ER/PR positive, HER2 negative, and low Ki-67 expression), Luminal B (HER2-) (i.e. ER/PR positive, HER2 negative, and high Ki-67 expression), Luminal B (HER2+) (i.e. ER/PR positive, HER2 positive, and any Ki-67 level), HER2-enriched (i.e. ER and PR negative, HER2 positive), and TNBC (i.e. ER, PR, and HER2 negative).

Statistical analysis

This was an exploratory study without a formal a priori sample size calculation. The statistical analysis was carried out using SPSS software (Version 26.0, IBM, https://www.ibm.com), MedCalc software (Version 22.0, MedCalc, https://www.medcalc.org) and R software (version 4.3.1, https://www.r-project.org). Measurements with normal distribution were expressed as the means ± standard deviations (SD), and independent t-test was used for comparison between two groups. Non-normally distributed metric data were represented by median (interquartile range), and differences in various indicators between two groups were analyzed using Mann-Whitney U test. Categorical variables were expressed as counts (percentages), and comparisons of statistical data were made using the chi-square test or the Fisher’s test.

Receiver operating characteristic (ROC) curve analysis and area under the curve (AUC) calculation were used to assess the performance of a binary classifier system and the diagnostic capability of the model. Univariate and multivariate logistic regression analyses were performed to identify the independent factors distinguishing breast cancer subtypes and to establish single and joint models. Variables with a P < 0.1 from the univariate regression analysis were included in the multivariate regression analysis. The DeLong test was employed to compare the differences in ROC curves among different models. To address internal validation and variable selection validation, LASSO regression with 10‑fold cross‑validation and bootstrap‑based optimism correction (200 resamples) was performed for all predictive models. The following R packages were used: glmnet, pROC, and caret. Statistical significance was defined as P < 0.05.

Results

Population characteristics

Of the 280 patients who underwent breast SyMRI, 164 were excluded and 116 patients (mean age, 48.64 ± 10.86 years; age range, 26–74 years) with 118 breast cancer lesions (mean diameter, 25.43 ± 15.36 mm; diameter range, 11–78 mm) met the inclusion criteria. Two patients had bilateral lesions. Of them, 93 (78.8%) were invasive ductal carcinomas (IDCs), five (4.2%) were invasive lobular carcinomas, five (4.2%) were invasive papillary carcinomas, three (2.5%) were ductal carcinomas in situ, three (2.5%) were tubular carcinomas, two (1.7%) were metaplastic carcinomas, two (1.7%) were encapsulated papillary carcinomas, one (0.8%) was carcinoma with medullary features, one (0.8%) was lobular carcinoma in situ, one (0.8%) was solid papillary carcinoma, one (0.8%) was cribriform carcinoma, and one (0.8%) was mucinous carcinoma.

Histopathology

Classification of the tumors by histopathological analysis was as follows: 99 (83.9%) were ER positive and 19 (16.1%) were ER negative; 96 (81.4%) were PR positive and 22 (18.6%) were PR negative; 31 (26.3%) were HER2 positive and 87 (73.7%) were HER2 negative; 103 (87.3%) tumors were Ki-67 high proliferation and 15 (12.7%) were Ki-67 low proliferation. Based on prognostic factor surrogates, 100 (84.8%) were classified as luminal tumors (15 were luminal A, 24 were luminal B-HER2 positive, and 61 were luminal B-HER2 negative), 7 (5.9%) were HER2 enriched, and 11 (9.3%) were TNBC.

Quantitative parameters of receptor status and proliferation rate

Representative images are shown in Fig. 1. Inter-observer and intra-observer ICCs for all quantitative parameters were > 0.75, indicating excellent agreement. The quantitative parametric values of SyMRI and MUSE DWI, based on receptor status (ER, PR, and HER2) and Ki-67, are presented in Table 1. ER, PR, and HER2 status, as well as Ki-67, all showed significantly different quantitative values for SyMRI. ER-positive tumors exhibited significantly lower values of T2-Pre, ∆T2, and T2-ratio compared to ER-negative tumors (P = 0.001, P = 0.009, and P = 0.023, respectively). Likewise, PR-positive tumors displayed significantly lower values of T2-Pre, T2-Gd, and ∆T2 than PR-negative tumors (P = 0.003, P = 0.028, and P = 0.023, respectively). HER2-positive tumors showed significantly higher T1-Pre, ∆T1, and T1-ratio values than HER2-negative tumors (n = 87) (P = 0.003, P = 0.001 and P = 0.012, respectively). Ki-67 high-proliferation tumors (n = 107) had significantly higher T1-Pre, ∆T1, and T1-ratio (P = 0.001, P = 0.002, and P = 0.014, respectively) and lower PD-Pre (P = 0.022) than Ki-67 low-proliferation tumors (n = 15). In terms of ADC values, no significant differences were observed based on ER, PR, HER2, or Ki-67. Table 2 demonstrates the effectiveness of SyMRI in predicting the expression of IHC marker expression.

Fig. 1.

Example of MUSE ADC map and synthetic MRI parameters in a 57-year-old female with invasive ductal carcinoma of the left breast (luminal B subtype: ER+, PR+, HER2–, Ki-67: 30%). (a) MUSE ADC map, (b) ADC map, (c) T1-Pre map, (d) T1-Gd map, (e) T2-Pre map, (f) T2-Gd map, (g) PD-Pre map, and (h) PD-Gd map. ROIs were placed in the tumors shown on the synthetic images. MRI quantitative parameter values of 0.885 × 10^− 3mm²/s (ADC), 1778.57ms (T1-Pre), 894.028ms (T1-Gd), 79.0691ms (T2-Pre), 78.0088ms (T2-Gd), 77.0245 (PD-Pre) and 81.216 (PD-Gd), respectively. ADC apparent diffusion coefficient, MUSE multiple sensitivity encoding.

Table 1.

MRI quantitative parameters stratified by immunohistochemistry (ICH) of breast cancer.

Variable	ER (+) (n = 99)	ER (-) (n = 19)	P	PR (+) (n = 96)	PR (-) (n = 22)	P	HER2 (+) (n = 31)	HER2 (-) (n = 87)	P	Ki-67>14% (n = 103)	Ki-67 ≤ 14% (n = 15)	P
ADC value (10− 3mm2/s)	0.965 ± 0.107	0.992 ± 0.112	0.332	0.953 (0.895,1.023)	0.982 (0.887,1.068)	0.425	0.954 (0.848,1.049)	0.954 (0.911,1.023)	0.633	0.954 (0.888,1.023)	0.954 (0.917,1.148)	0.238
T1-Pre (ms)	1255.51 ± 259.62	1299.40 ± 289.63	0.509	1251.53 ± 261.03	1310.79 ± 276.91	0.344	1360.26 ± 167.85	1227.77 ± 283.25	0.003	1292.58 ± 249.86	1056.58 ± 273.76	0.001
T1-Gd (ms)	569.46 (487.77,670.66)	616.59 (463.57,780.51)	0.351	566.40 (489.33,669.35)	610.40 (463.53,746.62)	0.415	576.74 (516.49,662.44,)	569.46 (480.45,698.77)	0.757	568.50 (494.02,679.70)	575.32 (465.88,718.72)	0.888
∆T1 (ms)	676.88 (478.46,849.52)	666.91 (435.13,832.33)	0.801	671.72 (439.51,848.10)	700.05 (570.81,841.76)	0.585	765.98 ± 185.41	604.20 ± 305.54	0.001	705.56 (518.16,851.87)	498.38 (341.55,602.61)	0.002
T1-ratio	0.52 (0.44,0.62)	0.53 (0.43,0.56)	0.503	0.52 (0.43,0.62)	0.53 (0.48,0.56)	0.887	0.56 (0.52,0.62)	0.51 (0.41,0.59)	0.012	0.54 (0.46,0.62)	0.44 (0.31,0.52)	0.014
T2-Pre (ms)	73.75 ± 11.86	84.21 ± 13.17	0.001	71.96 (65.04,82.42)	80.51 (73.80,96.76)	0.003	76.20 ± 11.04	75.16 ± 13.19	0.697	76.13 ± 12.60	70.64 ± 12.09	0.116
T2-Gd (ms)	68.83 (59.49,77.18)	73.44 (65.26,83.84)	0.066	68.82 (59.38,76.55)	71.44 (66.14,84.00)	0.028	68.94 (61.60,78.10)	69.21 (60.86,78.01)	0.963	69.21 (61.60,78.10)	65.16 (59.33,70.95)	0.303
∆T2 (ms)	5.45 (0.23,10.18)	11.34 (5.45,15.14)	0.009	5.43 (0.38,10.20)	9.04 (6.00,14.06)	0.023	5.45 (2.64,9.96)	6.94 (0.23,12.12)	0.886	6.25 (2.08,11.34)	1.55 (−4.21,8.42)	0.100
T2-ratio	0.07 (0.01,0.13)	0.12 (0.07,0.18)	0.023	0.12 (0.07,0.17)	0.07 (0.01,0.13)	0.053	0.07 (0.03,0.12)	0.09 (0.01,0.15)	0.922	0.08 (0.03,0.14)	0.02 (−0.07,0.12)	0.120
PD-Pre	72.16 ± 14.90	74.03 ± 17.29	0.625	72.61 ± 14.75	71.77 ± 17.60	0.934	72.06 ± 12.96	72.60 ± 16.05	0.866	71.23 ± 14.25	80.85 ± 19.41	0.022
PD-Gd	81.22 (72.23,98.20)	82.16 (73.79,97.02)	0.829	80.71 (72.49,97.53)	82.96 (73.17,97.85)	0.804	82.41 ± 15.38	84.59 ± 18.37	0.555	80.20 (71.32,92.02)	92.90 (78.16,109.50)	0.050
∆PD	−11.96 (−19.44, −4.68)	−11.34 (−17.82, −2.92)	0.652	−11.49 ± 11.52	−11.89 ± 11.60	0.882	−10.35 ± 9.98	−11.99 ± 11.99	0.497	−11.46 ± 11.04	−12.23 ± 14.59	0.811
PD-ratio	−0.18 (−0.29, −0.07)	−0.18 (−0.27, −0.03)	0.621	−0.17 (−0.28, −0.06)	−018 (−0.27,0.08)	0.689	−0.17 (−0.29, −0.07)	−0.18 (−0.27, −0.07)	0.680	−0.18 (−0.27,−0.07)	−0.17 (−0.29,−0.02)	0.731

*ER estrogen receptor, PR progesterone receptor, HER2 human epidermal growth factor receptor 2, ADC apparent diffusion coefficient, ∆T1 T1-Pre – T1-Gd, ∆T2 T2-Pre – T2-Gd, ∆PD PD-Pre –PD-Gd, T1-ratio T1-Pre – T1-Gd/T1-Pre, T2-ratio T2-Pre – T2-Gd/T2-Pre, PD-ratio PD-Pre –PD-Gd/PD-Pre.

Table 2.

MRI quantitative parameter’s ability to predict the expression of IHC.

Prognostic factor	Synthetic parameter	AUC (95% CI)	Cut-Off	Sensitivity (%)	Specificity (%)	PPV (%)	NPV (%)	P
ER	T2-Pre	0.712 (0.622–0.792)	≤ 88.3116	88.89	47.37	89.80	45.00	0.002
	∆T2	0.691 (0.599–0.772)	≤ 10.9114	80.81	52.63	89.89	34.48	0.003
	T2-ratio	0.665 (0.572–0.749)	≤ 0.0604	46.46	89.47	95.83	24.28	0.009
PR	T2-Pre	0.704 (0.613–0.784)	≤ 72.1513	52.08	86.36	94.33	29.22	0.001
	T2-Gd	0.651 (0.557–0.736)	≤ 79.8447	87.50	45.45	87.50	45.45	0.016
	∆T2	0.656 (0.563–0.741)	≤ 6.0204	55.21	77.27	91.38	28.33	0.0168
HER2	T1 -Pre	0.648 (0.555–0.734)	> 1137.31	93.55	36.78	34.52	94.12	0.004
	∆T1	0.679 (0.586–0.762)	> 589.889	93.55	48.28	39.19	95.46	<0.001
	T1 ratio	0.653 (0.560–0.738)	> 0.452	96.77	37.93	35.71	97.06	0.0031
Ki-67	T1 -Pre	0.764 (0.677–0.838)	> 1180.45	72.82	86.67	97.40	31.71	<0.001
	∆T1	0.742 (0.654–0.818)	> 609.67	62.14	93.33	98.46	26.41	<0.001
	T1-ratio	0.697 (0.606–0.779)	> 0.451	77.67	66.67	94.12	30.30	0.015
	PD-Pre	0.650 (0.557–0.736)	≤ 82.1209	80.58	53.33	92.22	28.56	0.095

*AUC area under the curve, CI confidence interval, NPV negative predictive value, PPV positive predictive value.

Univariate and multivariate logistic regression analysis of molecular subtypes

The results of univariate and multivariate analyses of breast cancer stratified by molecular subtypes are summarized in Table 3. In the univariate regression analysis, T2-Pre, T2-Gd, enhancement, and histologic grade were associated with luminal subtype. T1-Gd, T1-ratio, T2-Pre, T2-Gd, ∆T2 and enhancement were associated with TNBC subtype. ADC values of MUSE DWI and histologic grade were associated with HER2-enriched subtype. In the multivariable regression analysis, T2-Pre (P = 0.037) and histologic grade (P < 0.001) were independent indicators of luminal tumors. T2-Pre (P < 0.001) and enhancement (P = 0.0445) were independent indicators of TNBC. Additionally, ADC values of MUSE DWI (P = 0.022) and histologic grade (P = 0.027) were identified as independent indicators of HER2-enriched tumors. Bootstrap optimism correction yielded corrected AUCs of 0.876 (luminal), 0.805 (HER2-enriched), and 0.832 (TNBC). LASSO selected both T2‑Pre and histologic grade for the luminal model (AUC = 0.876). For the HER2-enriched (n = 7) and TNBC (n = 11) models, LASSO selected no variables due to small sample sizes. However, the variables identified by univariate analysis (P < 0.1) were retained for further analysis.

Table 3.

Compare the MRI quantitative parameters, imaging features, and clinical-pathological characteristics of breast cancer lesions of different molecular subtypes.

	Lumanal (n = 100)	Non-Lumanal (n = 18)	P value (univariate analysis)	P value (multivariate analysis)	HER2-enriched (n = 7)	Non-HER2-enriched (n = 111)	P value (univariate analysis)	P value (multivariate analysis)	TNBC (n = 11)	Non-TNBC (n = 107)	P value (univariate analysis)	P value (multivariate analysis)
Mean age, years	48.76 ± 10.92	48.00 ± 10.83	0.784		51.00 ± 109.92	48.50 ± 10.94	0.554		46.09 ± 11.41	48.91 ± 10.83	0.413
Mean tumor size (mm)	25.24 ± 16.22	26.50 ± 9.47	0.748		24.86 ± 7.15	25.47 ± 15.75	0.918		27.55 ± 10.89	25.21 ± 15.77	0.632
Tumor shape (round or oval/lobulated/irregular)	18(18)/53(53)/29(29)	3(16.7)/8(44.4)/7(38.9)	0.522		1(14.3)/3(42.9)/3(42.9)	20(18.0)/58(52.3)/33(29.7)	0.529		2(18.2)/5(45.5)/4(36.4)	19(17.8)/56(52.3)/32(29.9)	0.781
Tumor margin (spiculated/non-spiculated)	26(26)/74(74)	5(27.8)/13(72.2)	0.875		2(28.6)/5(71.4)	29(26.1)/82(73.9)	0.887		3(27.3)/8(72.7)	28(26.2)/79(73.8)	0.937
Tumor T2 signal intensity (low-iso/high)	70(70)/30(30)	15(83.3)/3(16.7)	0.255		6(85.7)/1(14.3)	79(71.2)/32(28.8)	0.420		9(81.8)/2(18.2)	76(71.0)/31(29.0)	0.454
Enhancement (rim/non-rim)	35(35)/65(65)	12(66.7)/6(33.3)	0.016		3(42.9)/4(57.1)	44(39.6)/67(60.4)	0.886		9(81.8)/2(18.2)	38(35.5)/69(64.5)	0.009	0.0445
TIC pattern (persistent/plateau/washout)	4(4)/20(20)/76(76)	2(11.1)/4(22.2)/12(66.7)	0.258		1(14.3)/2(28.6)/4(57.1)	5(4.5)/22(19.8)/84(75.7)	0.209		1(9.1)/2(18.2)/8(72.7)	5(4.7)/22(20.6)/80(74.8)	0.716
Axillary node metastasis (+/-)	51(51)/49(49)	8(44.4)/10(55.6)	0.722		3(42.9%)/4(57.1%)	54(48.6%)/57(51.4%)	0.767		5(45.5)/6(54.5)	52(48.6)/55(51.4)	0.845
Histologic type (ductal/lobular/other)	79(79)/6(6)/15(15)	16(88.9)/0(0)/2(11.1)	0.457		7(100)/0/0	88(79.3)/6(5.4)/17(15.3)	0.998		9(81.8)/0(0)/2(18.2)	86(80.4)/6(5.6)/15(14.0)	0.905
Histologic grade (1or2/3)	79(79)/21(21)	2(11.1)/16(88.9)	<0.001	<0.001	2(28.6)/5(71.4)	79(71.2)/32(28.8)	0.035	0.0265	0(0)/11(100)	81(75.7)/26(24.3)	0.996
ADC value (10− 3mm2/s)	0.965 ± 0.106	0.993 ± 0.115	0.310		1.064 ± 0.107	0.963 ± 0.105	0.024	0.0218	0.949 ± 0.099	0.972 ± 0.109	0.493
T1-Pre (ms)	1257.33 ± 258.95	1291.73 ± 296.04	0.609		1376.71 ± 240.52	1255.38 ± 264.61	0.240		1237.66 ± 325.59	1265.14 ± 258.42	0.741
T1-Gd (ms)	569.48 (489.33,672.40)	592.55 (463.04,746.62)	0.291		568.50 (461.45,698.34)	569.51 (487.77,682.07)	0.619		709.27 (463.57,820.09)	569.46 (487.77,672.98)	0.099
∆T1 (ms)	675.04 (468.96,848.10)	672.97 (535.33,841.76)	0.747		799.20 ± 182.37	637.09 ± 290.54	0.143		516.81 ± 402.56	660.06 ± 271.84	0.121
T1-ratio	0.52 (0.44,0.62)	0.54 (0.45,0.56)	0.431		0.57 (0.53,0.62)	0.52 (0.43,0.61)	0.198		0.50 (0.38,0.55)	0.53 (0.45,0.62)	0.070
T2-Pre (ms)	73.83 ± 11.83	84.36 ± 13.53	0.002	0.037	76.39 ± 12.60	75.37 ± 12.68	0.836		94.99 (73.94,99.80)	73.75 (65.34,82.54)	0.001	<0.001
T2-Gd (ms)	67.69 ± 13.01	73.54 ± 9.74	0.068		70.47 ± 10.00	68.46 ± 12.89	0.683		75.50 ± 9.52	67.87 ± 12.82	0.052
∆T2 (ms)	5.45 (0.38,10.16)	12.08 (6.55,15.36)	0.167		5.39 (−2.60,13.33)	6.23 (1.06,10.62)	0.840		13.17 (8.16,19.22)	5.45 (0.23,10.21)	0.072
T2-ratio	0.07 (0.01,0.13)	0.13 (0.08,0.19)	0.238		0.07 (−0.04,0.17)	0.08 (0.02,0.14)	0.892		0.13 (0.12,0.19)	0.07 (0.01,0.13)	0.119
PD-Pre	72.15 ± 14.82	74.14 ± 17.78	0.609		79.17 ± 13.27	72.03 ± 15.32	0.231		70.94 ± 20.07	72.61 ± 14.78	0.728
PD-Gd	80.71 (71.55,97.53)	82.96 (75.10,97.85)	0.932		87.27 (73.79,103.66)	81.22 (72.23,95.53)	0.308		81.35 (75.54,91.46)	82.13 (72.23,98.47)	0.467
∆PD	−11.81 ± 11.74	−10.20 ± 10.12	0.584		10.31 ± 4.64	−11.45 ± 6.76	0.979		−9.40 ± 12.03	−11.78 ± 11.50	0.512
PD-ratio	−0.18 ± 0.18	−0.17 ± 0.17	0.816		−11.57 ± 11.74	−0.14 ± 0.08	0.612		−0.18 ± 0.21	−0.17 ± 0.17	0.902

*TIC time-signal intensity curve.

Combination of MRI and clinical-pathological parameters for discriminating molecular subtypes

The results of ROC analyses are shown in Table 4; Fig. 2. Incorporating both MRI and clinical-pathological parameters associated with each molecular subtype, joint parameters were established. The joint parameters demonstrated superior discriminatory performance, surpassing the AUC of any individual quantitative parameter. Delong tests revealed that the diagnostic efficiency of the joint parameters (T2-Pre + histologic grade) significantly outperformed that of T2-Pre alone for distinguishing the luminal subtype (P = 0.002). The diagnostic efficiency of the joint parameters (T2-Pre + enhancement) in differentiating between TNBC and non-TNBC was significantly superior to that of enhancement alone (P = 0.0492), although it was not significantly better than that of T2-Pre alone (P = 0.3162). When distinguishing HER2-enriched subtypes from other subtypes, the diagnostic efficiency of the joint parameters (ADC values of MUSE DWI + histologic grade) did not show a significant difference compared to that of ADC alone (P = 0.3855).

Table 4.

Diagnostic performances for the Prediction of Molecular Subtypes Using combined MRI quantitative parameters with clinical-pathological models.

Molecular subtype	MRI quantitative parameter	AUC (95% CI)	Cut-Off	Sensitivity (%)	Specificity (%)	PPV (%)	NPV (%)	P
Luminal vs. Non- luminal	T2-Pre	0.709 (0.619–0.789)	≤ 88.3116	89.00	50.00	90.8191	44.9919	0.0040
	Histologic grade	0.839 (0.761–0.901)	≤ 1	79.00	88.89	97.5319	43.2355	< 0.0001
	T2-Pre + Histologic grade	0.878 (0.805–0.931)	> 0.709728393	86.00	83.33	96.6296	51.7150	< 0.0001
HER2-enriched vs. Non- HER2-enriched	ADC value	0.760 (0.673–0.834)	> 947	100.00	48.65	10.9339	100.0000	0.0008
	Histologic grade	0.713 (0.622–0.793)	> 1	71.43	71.17	97.5188	13.5720	0.0245
	ADC value + Histologic grade	0.824 (0.743–0.888)	> 0.050897317	85.71	73.87	17.1344	98.7952	0.0005
TNBC vs. Non- TNBC	T2-Pre	0.814 (0.732–0.880)	> 91.6127	63.64	93.46	50.0033	96.1552	< 0.0001
	Enhancement	0.732 (0.642–0.809)	> 0	81.82	64.49	19.1473	97.1842	0.0004
	T2-Pre + Enhancement	0.854 (0.777–0.912)	> 0.033633525	100.00	60.75	20.7517	100.00	< 0.0001

Fig. 2.

ROC curves for single and combined parameters used to distinguish between luminal and non-luminal (a), HER2-enriched and non-HER2-enriched (b), and TN and non-TN (c). For luminal subtype (a), the joint model (T2-Pre + histologic grade) achieved the highest AUC (0.878), compared to T2-Pre alone (0.709) and histologic grade alone (0.839). For HER2-enriched subtype (b), the joint model (ADC + histologic grade) achieved an AUC of 0.824, similar to ADC alone (0.760). For TNBC (c), the joint model (T2-Pre + enhancement) achieved an AUC of 0.854, higher than enhancement alone (0.732). ROC receiver operating characteristic, ADC apparent diffusion coefficient, HER2 human epidermal growth factor receptor 2, TN triple-negative.

Discussion

We found that SyMRI parameters performed well in predicting prognostic factors and classifying molecular subtypes. While ADC value was only useful for distinguishing HER2-positive from HER2-negative tumors, SyMRI parameters were useful for distinguishing luminal from non-luminal and TNBC from non-TNBC, but not for HER2-enriched. In addition, SyMRI parameter T2-Pre was an independent predictor of luminal breast cancer and TNBC, and ADC value of MUSE DWI was an independent predictor for distinguishing HER2-enriched subtypes.

Breast cancer is a heterogeneous disease, and IHC reflects this heterogeneity. According to the expression of ER, PR, HER2 and Ki-67, breast cancer is divided into different molecular subtypes. These subtypes guide treatment and prognosis. Therefore, a non-invasive method reflecting the overall tumor state is urgently needed. SyMRI is a new technology that obtains multiple contrast-weighted images (T1, T2, PD) from a single scan and quantifies intrinsic relaxation times and proton density. It has been applied to breast cancer. We tried to use SyMRI parameters combined with ADC values to predict prognostic factors and identify molecular subtypes.

The results showed that SyMRI parameters and ADC values had significant effects, though some findings differed from previous studies. First, T2-Pre of ER- or PR- negative tumors was significantly higher than that of positive tumors, consistent with previous studies^28–32. We also found that ER- or PR- negative group had higher ∆T2 and T2-ratio values. Compared with other parameters, T2-Pre had the largest AUC in distinguishing ER or PR positive tumors. The difference in SyMRI parameter values is due to the varying fluid content within tumor cells. ER/PR-positive tumors typically have lower proliferation rates and lower cell density, which may manifest as matrix reaction and fibrosis, along with reduced neovascularization, leading to lower T2 values. Second, Gao et al. found that PD-Pre could distinguish HER2 positive tumors, while we used T1-Pre, ∆T1 and T1-ratio³⁰. In our study, T1-Pre, ∆T1, T1-ratio of HER2 positive tumors were higher than those of HER2 negative tumors, consistent with Li et al.³². In predicting Ki-67, some studies found that T2-related quantitative values were significantly different between groups^28,30, but our study did not replicate this finding. We found that T1 and PD parameters were more meaningful for the expression of Ki-67, and the Ki-67 high-expression group had higher T1-Pre, ∆T1, T1-ratio and lower PD-Pre. Previous studies have shown that SyMRI T1 values are highly correlated with Ki-67 expression^29,33,34. The T1-related values were higher in the high-expression group than in the low-expression group. This result was consistent with us but not with Li et al.³². In their study, T1 and PD in Ki-67 low-proliferation tumors were significantly higher than those in Ki-67 high-proliferation group. Due to partial volume effect, we excluded tumors with a lesion diameter of less than 10 mm, while Li’s study included tumors with 5–10 mm. The partial volume effect of surrounding tissues such as fat and glands may have affected Li et al. Few studies have reported PD for the evaluation of malignant tumors, these results require further confirmation.

Currently, there is no consensus on ADC value and IHC marker expression. Some studies have found that ADC values are related to IHC marker expression, with lower ADC values in ER/PR-positive breast cancer^13,14,35. However, we found no significant difference in ADC values for prognostic factors, aligning with Congru Yuan¹². This discrepancy may stem from variations in measurement techniques and cohort compositions¹³. Therefore, further verification is required for ADC values.

In molecular subtype classification, our results also differed from other studies. First, only T2-Pre was associated with luminal breast cancers. In contrast, Du et al. reported T2-Pre and ADC, and Gao et al. reported T1-Pre, T2-Pre, and T2-Gd as significant predictors^29,30. These differences may be due to imbalance between luminal and non-luminal groups. In this study, T2-Pre was higher in non-luminal than luminal cancers, consistent with Gao³⁰. The diagnostic ability of our SyMRI (AUC = 0.709) was higher than that of Gao et al. (AUC = 0.669). We also found that logistic regression combining SyMRI with histologic grade showed excellent diagnostic performance, with an AUC value of 0.878, which was higher than the joint model combining SyMRI, ADC and volume by Du et al. (AUC = 0.765). In distinguishing TNBC from non-TNBC, Amano et al. found that the mean T2 of TNBC was significantly higher than that of non-TNBC²⁸, which was consistent with our results. Previous studies indicate that TNBC exhibits higher levels of vascular endothelial growth factor, promoting angiogenesis^36,37. Additionally, TNBC demonstrates a higher microvessel density, which may contribute to higher T2 values. However, except for T2-Pre, we did not find other SyMRI parameters associated with TNBC, which was different from other reports^29,30,38,39. Multivariate regression analysis showed that T2-Pre was associated with TNBC. The joint model combining SyMRI with enhancement (AUC = 0.854) was comparable to that of Matsuda et al. (AUC = 0.858)³⁸, and higher than those of Du et al. and Zhang et al. (AUC = 0.742 and 0.709, respectively)^29,39. Gao et al. found that neither ADC nor SyMRI parameters could distinguish HER2-enriched from non-HER2-enriched³⁰. Du et al. proposed that ADC value was associated with HER2-enriched tumors, and there are no SyMRI parameters that can differentiate between HER2-enriched and non-HER2-enriched tumors²⁹, which was consistent with our results. Multiple regression analysis revealed that the joint model of ADC and histologic grade demonstrated the highest diagnostic efficacy (AUC = 0.824), surpassing the combined model proposed by Du et al. (AUC = 0.736).

Our study has several limitations. First, it is a single-center study with a limited sample size, particularly for HER2‑enriched (n = 7) and TNBC (n = 11) subtypes, which limits statistical power and subgroup robustness. For these subtypes, the very small sample sizes raise concerns regarding overfitting and limit clinical applicability. Therefore, these results should be interpreted with caution. Second, we excluded lesions < 10 mm and non‑mass lesions to minimize partial volume effects and ensure ROI consistency, which may limit generalizability to smaller or non‑mass lesions. Future studies should specifically evaluate these excluded populations. Third, we did not perform whole‑lesion volumetric or radiomics‑based analysis. Future studies using volumetric or radiomics approaches are warranted. Fourth, we lacked external validation and comparison with existing radiomics or machine learning approaches. Therefore, these findings are preliminary and require validation in larger multicenter cohorts.

Conclusion

In conclusion, quantitative parameters derived from SyMRI and MUSE DWI represent a promising approach for differentiating receptor status, proliferation rate, and molecular subtypes in breast cancer patients. However, these findings are preliminary due to the limited sample size, underrepresentation of HER2‑enriched and TNBC subtypes, and lack of external validation. Multicenter prospective studies with larger cohorts are required to validate these results.

Author contributions

Xiufeng Song led the project. Lei Yang wrote the manuscript. Haifei Lan processed the data and wrote the part of the manuscript. Yang Li reviewed the article. Zaixian Zhang, Xiaowen Xu, Na Zhu, and Shuhui Wang were mainly involved in sample collection and collation.

Funding

None.

Data availability

The datasets used and/or analysed during the current study available from the corresponding author on reasonable request.

Declarations

Competing interests

The authors declare no competing interests.