Implementing NMS-Free Training Using YOLOv10 on Mammographic Images to Detect Breast Cancer

This study focuses on developing an automated, data-driven model for detecting breast cancer in mammograms, aimed at assisting physicians in their decision-making processes during breast cancer screening and detection programs. The research utilizes publicly available datasets, to apply Deep learning techniques on a proprietary dataset of full-field digital mammography. For this investigation, YOLOv10 architecture is employed for detecting breast cancer in mammograms and compared all the variants of this model. Despite the challenges posed by difficult-to-detect anomalies like asymmetries and distortions in the proprietary dataset, the YOLOv10 model demonstrated enhanced detection capabilities compared to prior architectures. However, it is crucial to treat these outputs as qualitative tools that require clinical radiologic evaluation. Overall, this study focuses on determining the best model which serves as a reliable predictive system to aid cognitive processes and decision-making, promoting its potential adoption in clinical practice.