Efficient Thermography Guided Learning for Breast Cancer Detection

Abstract

Early-stage breast cancer detection is often thwarted due to privacy concerns, the need for regular scanning, among other factors, thereby severely reducing the survival rate of patients. Thermography is an emerging low cost, portable, non-invasive, and privacy-sensitive technique for early-stage breast cancer detection gaining popularity over the traditional mammography based technique that requires expert intervention in a lab setup. Earlier proposals for machine learning augmented thermography for early-stage breast cancer detection suffer from precision as well as performance challenges. We developed a novel voting based machine learning model with on the fly parallel retraining using the Dask library. Experimental evaluation reveals that our novel high-performance thermography based learning technique brings up the accuracy of early-stage life-saving breast cancer detection to 93%.