Noninvasive identification of HER2-low-positive status by MRI-based deep learning radiomics predicts the disease-free survival of patients with breast cancer.

OBJECTIVE: This study aimed to establish a MRI-based deep learning radiomics (DLR) signature to predict the human epidermal growth factor receptor 2 (HER2)-low-positive status and further verified the difference in prognosis by the DLR model. METHODS: A total of 481 patients with breast cancer who underwent preoperative MRI were retrospectively recruited from two institutions. Traditional radiomics features and deep semantic segmentation feature-based radiomics (DSFR) features were extracted from segmented tumors to construct models separately. Then, the DLR model was constructed to assess the HER2 status by averaging the output probabilities of the two models. Finally, a Kaplan‒Meier survival analysis was conducted to explore the disease-free survival (DFS) in patients with HER2-low-positive status. The multivariate Cox proportional hazard model was constructed to further determine the factors associated with DFS. RESULTS: First, the DLR model distinguished between HER2-negative and HER2-overexpressing patients with AUCs of 0.868 and 0.763 in the training and validation cohorts, respectively. Furthermore, the DLR model distinguished between HER2-low-positive and HER2-zero patients with AUCs of 0.855 and 0.750, respectively. Cox regression analysis showed that the prediction score obtained using the DLR model (HR, 0.175; p = 0.024) and lesion size (HR, 1.043; p = 0.009) were significant, independent predictors of DFS. CONCLUSIONS: We successfully constructed a DLR model based on MRI to noninvasively evaluate the HER2 status and further revealed prospects for predicting the DFS of patients with HER2-low-positive status. CLINICAL RELEVANCE STATEMENT: The MRI-based DLR model could noninvasively identify HER2-low-positive status, which is considered a novel prognostic predictor and therapeutic target. KEY POINTS: • The DLR model effectively distinguished the HER2 status of breast cancer patients, especially the HER2-low-positive status. • The DLR model was better than the traditional radiomics model or DSFR model in distinguishing HER2 expression. • The prediction score obtained using the model and lesion size were significant independent predictors of DFS.

Multiparametric MRI Radiomics With Machine Learning for Differentiating HER2-Zero, -Low, and -Positive Breast Cancer: Model Development, Testing, and Interpretability Analysis.

BACKGROUND: MRI radiomics has been explored for three-tiered classification of breast cancer HER2 expression (i.e., HER2-zero, HER2-low, or HER2-positive), although understanding of how such models reach their predictions is lacking. OBJECTIVE: To develop and test multiparametric MRI radiomics machine-learning models for differentiating three-tiered HER2 expression levels in patients with breast cancer, and to explain the contributions of model features through local and global interpretations using SHapley Additive exPlanation (SHAP) analysis. METHODS: This retrospective study included 737 patients (mean age, 54.1±10.6 years) with breast cancer from two centers (center 1: n=578; center 2: n=159), who underwent breast MRI and had HER2 expression determined after excisional biopsy. Analysis entailed two tasks: differentiating HER2-negative (i.e., HER2-zero or HER2-low) from HER2-positive tumors (task 1), and differentiating HER2-zero from HER2-low tumors (task 2). For each task, patients from center 1 were randomly assigned in 7:3 ratio to training (task 1: n=405; task 2: n=284) or internal test (task 1: n=173; task 2: n=122) sets; those from center 2 formed an external test set (task 1: n=159; task 2: n=105). Radiomics features were extracted from early-phase dynamic contrast-enhanced images (DCE), T2-weighted images (T2WI), and DWI. For each task, a support vector machine (SVM) was used for feature selection; a multiparametric radiomics score (radscore) was computed using feature weights from SVM correlation coefficients; conventional MRI and combined models were constructed; and model performances were evaluated. SHAP analysis was used to provide local and global interpretations for model outputs. RESULTS: In the external test set, for task 1, AUCs for the conventional MRI model, radscore, and combined model were 0.624, 0.757, and 0.762, respectively; for task 2, AUC for radscore was 0.754, and no conventional MRI model or combined model could be constructed. SHAP analysis identified early-phase DCE features as having the strongest influence for both tasks; T2WI features also had a prominent role for task 2. CONCLUSION: The findings indicate suboptimal performance of MRI radiomics models for noninvasive characterization of HER2 expression. CLINICAL IMPACT: The study provides an example of the use of SHAP interpretation analysis to better understand predictions of imaging-based machine learning models.

Abbreviated Versus Full-Protocol MRI for Breast Cancer Neoadjuvant Chemotherapy Response Assessment: Diagnostic Performance by General and Breast Radiologists.

BACKGROUND. Abbreviated protocols could allow wider adoption of MRI in patients undergoing breast cancer neoadjuvant chemotherapy (NAC). However, abbreviated MRI has been explored primarily in screening settings. OBJECTIVE. The purpose of this article was to compare diagnostic performance of abbreviated MRI and full-protocol MRI for evaluation of breast cancer NAC response, stratifying by radiologists' breast imaging expertise. METHODS. This retrospective study included 203 patients with breast cancer (mean age, 52.1 ± 11.2 [SD] years) from two hospitals who underwent MRI before NAC initiation and after NAC completion before surgical resection from March 2017 to April 2021. Abbreviated MRI was extracted from full-protocol MRI and included the axial T2-weighted sequence and precontrast and single early postcontrast T1-weighted sequences. Three general radiologists and three breast radiologists independently interpreted abbreviated and full-protocol MRI in separate sessions, identifying enhancing lesions to indicate residual tumor and measuring lesion size. The reference standard was presence and size of residual tumor on pathologic assessment of post-NAC surgical specimens. RESULTS. A total of 50 of 203 patients had pathologic complete response (pCR). Intraobserver and interobserver agreement for abbreviated and full-protocol MRI for general and breast radiologists ranged from substantial to nearly perfect (κ = 0.70-0.81). Abbreviated MRI compared with full-protocol MRI showed no significant difference for general radiologists in sensitivity (54.7% vs 57.3%, p > .99), specificity (92.8% vs 95.6%, p = .29), or accuracy (83.4% vs 86.2%, p = .30), nor for breast radiologists in sensitivity (60.0% vs 61.3%, p > .99), specificity (94.6% vs 97.4%, p = .22), or accuracy (86.0% vs 88.5%, p = .30). Sensitivity, specificity, and accuracy were not significantly different between protocols for any reader individually (p > .05). Mean difference in residual tumor size on MRI relative to pathology for abbreviated protocol ranged for general radiologists from -0.19 to 0.03 mm and for breast radiologists from -0.15 to -0.05 mm, and for full protocol ranged for general radiologists from 0.57 to 0.65 mm and for breast radiologists from 0.66 to 0.79 mm. CONCLUSION. Abbreviated compared with full-protocol MRI showed similar intraobserver and interobserver agreement and no significant difference in diagnostic performance. Full-protocol MRI but not abbreviated MRI slightly overestimated pathologic tumor sizes. CLINICAL IMPACT. Abbreviated protocols may facilitate use of MRI for post-NAC response assessment by general and breast radiologists.

Evaluation of the Effects of Anti-PD-1 Therapy on Triple-Negative Breast Cancer in Mice by Diffusion Kurtosis Imaging and Dynamic Contrast-Enhanced Imaging.

BACKGROUND: The monitoring of immunotherapies is still based on changes in the tumor size in imaging, with a long evaluation period and low sensitivity. PURPOSE: To investigate the effectiveness of diffusion kurtosis imaging (DKI) and dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) in assessing the therapeutic efficacy of anti-programmed death-1 (PD-1) therapy in a mouse triple negative breast cancer (TNBC) model. STUDY TYPE: Prospective. ANIMAL MODEL: A total of 54 BALB/c mouse subcutaneous 4 T1 transplantation models of TNBC. FIELD STRENGTH/SEQUENCE: A 3.0-T; turbo spin echo (TSE) T2-weighted imaging, DKI with seven b values (0, 500, 1000, 1500, 2000, 2500, and 3000 sec/mm2 ) and T1-twist DCE acquisition series. ASSESSMENT: DKI and DCE-MRI parameters were evaluated by two radiologists independently. Regions of interest (ROIs) were drawn manually on the maximum cross-sectional area of the lesion; care was taken to avoid necrotic areas. The tumor cell density, the CD45 and CD31 levels were analyzed by two pathologists. STATISTICAL TESTS: The two-tailed unpaired t-test, Mann-Whitney U test, Fisher's exact test and Pearson correlation coefficient were performed. A P < 0.05 was considered statistically significant. RESULTS: The apparent diffusion coefficient (ADC), mean diffusivity (MD), Ktrans and Kep values were significantly different between the two groups at each time point after treatment. There were significant differences in the mean kurtosis (MK) and Ve values between the two groups at 5 and 10 days after treatment but no significant differences at 15 days (P = 0.317 and 0.183, respectively). The ADC and MD values were significantly correlated with tumor cell density (ADC, r = -0.833; MD, r = 0.890) and the CD45 level (ADC, r = 0.720; MD, r = 0.718). The Ktrans and Kep values were significantly correlated with the CD31 level (Ktrans , r = 0.820; Kep , r = 0.683). DATA CONCLUSION: DKI and DCE-MRI could reflect the changes in tumor microstructure and tumor tissue vasculature after anti-PD-1 therapy, respectively. LEVEL OF EVIDENCE: 1 TECHNICAL EFFICACY: Stage 4.

Performance of radiomics models for tumour-infiltrating lymphocyte (TIL) prediction in breast cancer: the role of the dynamic contrast-enhanced (DCE) MRI phase.

OBJECTIVE: To systematically investigate the effect of imaging features at different DCE-MRI phases to optimise a radiomics model based on DCE-MRI for the prediction of tumour-infiltrating lymphocyte (TIL) levels in breast cancer. MATERIALS AND METHODS: This study retrospectively collected 133 patients with pathologically proven breast cancer, including 73 patients with low TIL levels and 60 patients with high TIL levels. The volumes of breast cancer lesions were manually delineated on T2-weighted imaging (T2WI), diffusion-weighted imaging (DWI), and each phase of DCE-MRI, followed by 6250 quantitative feature extractions. The least absolute shrinkage and selection operator (LASSO) method was used to select predictive feature sets for the classifiers. Four models were developed for predicting TILs: (1) single enhanced phase radiomics models; (2) fusion enhanced multi-phase radiomics models; (3) fusion multi-sequence radiomics models; and (4) a combined radiomics-based clinical model. RESULTS: Image features extracted from the delayed phase MRI, especially DCE_Phase 6 (DCE_P6), demonstrated dominant predictive performances over features from other phases. The fusion multi-sequence radiomics model and combined radiomics-based clinical model achieved the highest predictive performances with areas under the curve (AUCs) of 0.934 and 0.950, respectively; however, the differences were not statistically significant. CONCLUSION: The DCE-MRI radiomics model, especially image features extracted from the delayed phases, can help improve the performance in predicting TILs. The radiomics nomogram is effective in predicting TILs in breast cancer. KEY POINTS: • Radiomics features extracted from DCE-MRI, especially delayed phase images, help predict TIL levels in breast cancer. • We developed a nomogram based on MRI to predict TILs in breast cancer that achieved the highest AUC of 0.950.

Nomogram for the prediction of triple-negative breast cancer histological heterogeneity based on multiparameter MRI features: A preliminary study including metaplastic carcinoma and non- metaplastic carcinoma.

Objectives: Triple-negative breast cancer (TNBC) is a heterogeneous disease, and different histological subtypes of TNBC have different clinicopathological features and prognoses. Therefore, this study aimed to establish a nomogram model to predict the histological heterogeneity of TNBC: including Metaplastic Carcinoma (MC) and Non-Metaplastic Carcinoma (NMC). Methods: We evaluated 117 patients who had pathologically confirmed TNBC between November 2016 and December 2020 and collected preoperative multiparameter MRI and clinicopathological data. The patients were randomly assigned to a training set and a validation set at a ratio of 3:1. Based on logistic regression analysis, we established a nomogram model to predict the histopathological subtype of TNBC. Nomogram performance was assessed with the area under the receiver operating characteristic curve (AUC), calibration curve and decision curve. According to the follow-up information, disease-free survival (DFS) survival curve was estimated using the Kaplan-Meier product-limit method. Results: Of the 117 TNBC patients, 29 patients had TNBC-MC (age range, 29-65 years; median age, 48.0 years), and 88 had TNBC-NMC (age range, 28-88 years; median age, 44.5 years). Multivariate logistic regression analysis demonstrated that lesion type (p = 0.001) and internal enhancement pattern (p = 0.001) were significantly predictive of TNBC subtypes in the training set. The nomogram incorporating these variables showed excellent discrimination power with an AUC of 0.849 (95% CI: 0.750-0.949) in the training set and 0.819 (95% CI: 0.693-0.946) in the validation set. Up to the cutoff date for this analysis, a total of 66 patients were enrolled in the prognostic analysis. Six of 14 TNBC-MC patients experienced recurrence, while 7 of 52 TNBC-NMC patients experienced recurrence. The DFS of the two subtypes was significantly different (p=0.035). Conclusions: In conclusion, we developed a nomogram consisting of lesion type and internal enhancement pattern, which showed good discrimination ability in predicting TNBC-MC and TNBC-NMC.