Triaging subjects with palpable breast masses for biopsy, follow-up or treatment using AI applied to breast ultrasound in a low-resource setting - A feasibility study

Breast cancer is more common in premenopausal women in developing countries. In the absence of screening, women present with symptoms, usually a palpable mass; however most palpable masses are benign. Automatic triaging of palpable breast masses using ultrasound and Artificial Intelligence (AI) in a low-resource setting would be beneficial by characterizing them (benign, suspicious, malignant) for follow-up with further imaging (suspicious) or biopsy (malignant). Further, automatic prediction of estrogen-receptor-positive (ER+) status of malignant lesions from ultrasound could allow initiation of primary endocrine therapy when appropriate and accelerate the need for specialized immunohistochemistry (IHC), if the tumor appears more likely to be a triple receptor negative, which is more aggressive and faster growing. In this work, we explore methods using Artificial Intelligence (AI) for automatic classification of breast lesions as benign, suspicious and malignant, and prediction of ER+ status for malignant lesions.