Automatic detection of breast masses using deep learning with YOLO approach

Introduction Breast cancer is the most common malignant tumor among women. Mammography is the specific type of X-ray recommended to examine the breasts. However, they are difficult to interpret due to the size of the lesions, shape, indefinite borders, and low contrast of the masses with respect to healthy tissue, mainly in very dense breasts. Computer-aided detection (CAD) systems increase the efficiency of diagnoses and reduce the workload of specialists. Purpose A CAD system that uses artificial intelligence (AI) based on "You Only Look Once" (YOLO), with two models YOLOv5x and YOLOv5s, is tested for the detection of breast nodules from mammography. Method Transfer learning and data augmentation techniques were applied. Image sets for training and validation were created from an international database (Vindr-Mammo). The network was trained and validated, and for the best model obtained, an external test was performed from a second database belonguing to “The Mammographic Image Analysis Society” (MIAS Database). Results The best model was obtained with YOLOv5x. This reached a maximum sensitivity of 80% in internal validation and 72% with external test data. Conclusion YOLOv5x and YOLOv5s models showed potential for the task of detecting masses from mammographies.