Distinguishing Molecular Subtypes of Breast Cancer Based on Computer-Aided Diagnosis of DCE-MRI

Shannon Agner¹, Mark Rosen², Sarah Englander², Diana Sobers¹, Kathleen Thomas², John Tomaszewski³, Michael Feldman³, Shridar Ganesan¹, Mitchell Schnall², Anant Madabhushi¹

Previous studies based on visual inspection of breast tumors suggest that molecular subtypes of breast cancer are associated with distinct imaging phenotypes on DCE-MRI. In this study, we develop a computer-aided diagnosis tool that utilizes textural kinetics, an attribute that captures time related changes in internal lesion texture, to distinguish between 20 triple negative (estrogen receptor (ER) negative/ progesterone receptor (PR) negative/ human epidermal growth factor (HER2) receptor negative) and 21 ER positive tumors. Our CAD system was found to outperform classifiers that were driven by morphology, signal intensity kinetics, peak contrast texture, and pharmacokinetic parameters.