A Comparative Study of Breast Mass Detection Using YOLOv8 Deep Learning Model in Various Data Scenarios on Multi-View Digital Mammograms

Abstract

Breast cancer is one of the most common types of cancer among women worldwide. It typically begins with abnormal cell growth in the breast glands or milk ducts and can spread to other tissues. Many breast cancer cases start with the presence of a mass and should be carefully examined. Masses can be monitored using X-ray-based digital mammography images, including right craniocaudal, left craniocaudal, right mediolateral oblique, and left mediolateral oblique views. In this study, automatic mass detection and localization were performed on mammography images taken from the full-field digital mammography VinDr-Mammo dataset using the YOLOv8 deep learning model. Three different scenarios were tested: raw data, data with preprocessing to crop breast regions, and data with only mass regions cropped to a 1.2x ratio. The data were divided into 80% for training and 10% each for validation and testing. The results were evaluated using performance metrics such as precision, recall, F1-score, mAP, and training graphs. At the end of the study, it is demonstrated that the YOLOv8 deep learning model provides successful results in mass detection and localization, indicating its potential use as a computer-based decision support system.