Towards a Cost-Effective Predictive Mammogram Classification Model for Breast Cancer Diagnosis

Abstract

Breast cancer is the deadliest common cancer in women and slightly in men worldwide. Routine mammography is the standard technique for preventive care, detection and classification of breast cancer before a biopsy. It has come to our attention that, routine mammography is still a manual process, prone to human errors which result in unnecessary costs on both the patient and medical institute which may lead to loss of life. In this paper, we developed a prototype cost-effective predictive mammogram classification model for breast cancer diagnosis using Deep Learning Studio performing data augmentation, transfer learning and careful data preprocessing. The resulting prototype model was trained on a publicly available In-breast dataset and achieve above human-level performance on the classification of mammograms. Finally, it is worth noting that the experiments we performed showed some degree of confidence that our prototype could improve the currently used methods for predictive mammogram classification.