Multiparametric MR Imaging Radiomics Signatures for Assessing the Recurrence Risk of ER+/HER2− Breast Cancer Quantified With 21‐Gene Recurrence Score

• Published in: Journal of magnetic resonance imaging Vol. 58; no. 2; pp. 444 - 453
• Main Authors: Chen, Yang, Tang, Wei, Liu, Wei, Li, Ruimin, Wang, Qifeng, Shen, Xigang, Gong, Jing, Gu, Yajia, Peng, Weijun
• Format: Journal Article
• Language: English
• Published: Hoboken, USA John Wiley & Sons, Inc 01.08.2023; Wiley Subscription Services, Inc
• Subjects: Breast cancer; breast neoplasms; Breast Neoplasms - diagnostic imaging; Breast Neoplasms - genetics; Cancer therapies; Diffusion coefficient; Diffusion Magnetic Resonance Imaging; ErbB-2 protein; Female; Field strength; Gene expression; Gene mapping; genomics; Health risks; Humans; image interpretation; Image processing; Image segmentation; Magnetic resonance imaging; Magnetic Resonance Imaging - methods; Medical imaging; Multiparametric Magnetic Resonance Imaging - methods; neoplasm recurrence; Patients; Radiomics; Retrospective Studies; Statistical analysis; Statistical tests; Support vector machines; Training; genomics; neoplasm recurrence; breast neoplasms; image interpretation
• ISSN: 1053-1807; 1522-2586; 1522-2586
• DOI: 10.1002/jmri.28547

Background While the Oncotype DX 21‐gene recurrence score (RS) has been recommended for guiding ER+/HER2− breast cancer treatment decisions, it is limited by cost and availability. Purpose To develop a multiparametric MRI‐based radiomics model for assessing ER+/HER2− breast cancer patients' 21‐gene RS. Study Type Retrospective. Subjects A total of 151 patients with pathologically confirmed ER+/HER2− breast cancers, who underwent preoperative breast MR examinations and 21‐gene expression assays, divided into training (n = 106) and validation (n = 45) cohorts. Field Strength/Sequence T2‐weighted imaging (T2WI), diffusion‐weighted imaging (DWI), and dynamic contrast‐enhancement (DCE) sequence at 1.5 T or 3 T. Assessment A total of 1046 radiomics features were extracted from each MRI sequence with a manual lesion segmentation method. After feature dimension reduction by the recursive feature elimination method and dataset balance by the synthetic minority oversampling technique, linear support vector machine classifier models were built to distinguish high RS (RS ≥ 26) from low RS (RS < 26) from T2WI, DWI apparent diffusion coefficient (ADC) maps, DCE and their combination (multiparametric). A model based on clinical characteristics and a fusion model combining clinical characteristics and multiparametric MRI were also built. Statistical Tests Receiver operating characteristic (ROC) curve analysis and De Long's test with Bonferroni correction were used. A P value <0.01 was considered statistically significant. Results The area under the ROC curve (AUC) value of multiparametric radiomics model was 0.92, significantly higher than DCE (0.83), T2WI (0.78), and ADC (0.77) models in the training cohort. The radiomics model also achieved good performance in the validation cohort (AUC = 0.77). The fusion model had significantly higher performance than the clinical model in both the training (AUC = 0.92 and 0.64, respectively) and validation cohorts (AUC = 0.78 and 0.62, respectively). Data Conclusion The proposed multiparametric MRI‐based radiomics models may have potential to help distinguish ER+/HER2− breast cancer patients' recurrence risk. Evidence Level 3. Technical Efficacy Stage 2.