Computational approach to radiogenomics of breast cancer: Luminal A and luminal B molecular subtypes are associated with imaging features on routine breast MRI extracted using computer vision algorithms.

PURPOSE: To identify associations between semiautomatically extracted MRI features and breast cancer molecular subtypes. METHODS: We analyzed routine clinical pre-operative breast MRIs from 275 breast cancer patients at a single institution in this retrospective, Institutional Review Board-approved study. Six fellowship-trained breast imagers reviewed the MRIs and annotated the cancers. Computer vision algorithms were then used to extract 56 imaging features from the cancers including morphologic, texture, and dynamic features. Surrogate markers (estrogen receptor [ER], progesterone receptor [PR], human epidermal growth factor receptor-2 [HER2]) were used to categorize tumors by molecular subtype: ER/PR+, HER2- (luminal A); ER/PR+, HER2+ (luminal B); ER/PR-, HER2+ (HER2); ER/PR/HER2- (basal). A multivariate analysis was used to determine associations between the imaging features and molecular subtype. RESULTS: The imaging features were associated with both luminal A (P = 0.0007) and luminal B (P = 0.0063) molecular subtypes. No association was found for either HER2 (P = 0.2465) or basal (P = 0.1014) molecular subtype and the imaging features. A P-value of 0.0125 (0.05/4) was considered significant. CONCLUSION: Luminal A and luminal B molecular subtype breast cancer are associated with semiautomatically extracted features from routine contrast enhanced breast MRI.

Duke Scholars

Author Lars Johannes L Grimm Radiology, Breast Imaging

Author Maciej A Mazurowski Radiology