Optimized Machine Learning Technique for Breast Cancer Detection and Classification

Abstract Breast cancer is one of the tumor types that causes the breast cells to become active causing mortality among females worldwide. It enables doctors to take appropriate action toward patients, it is imperative to identify and classify breast cancer in its early stages. Nevertheless, surgical biopsy, fine needle aspiration, and mammography are frequently employed techniques for diagnosis. These techniques provide low effectiveness and sensitivity in identifying breast cancer. Hence, the optimized You Look Only Once Version 3 (YOLOv3) technique trains the ultrasound image datasets to identify the breast tumor by extracting the features and classifying them as benign, malignant, and normal types. In this process, increasing the number of training datasets for the network decreases errors and boosts system performance followed by anticipating normal, benign, and cancerous tissues. The Wisconsin Breast Cancer Datasets (WBCD) ultrasonography pictures are employed for training and testing. The optimized YOLOv3 achieves accuracy, sensitivity, and specificity of 97.21%, 93.63%, and 96.81% and presents a promising system to support radiologists in clinical settings compared to existing techniques, and the proposed approach exhibits superior performance, underscoring its efficiency in classifying breast tumors as benign or malignant in real-time using computer-assisted diagnosis.