Autonomous Diagnosis System of Breast Cancer

This paper focuses on a study of medical image techniques using deep learning, specifically addressing methods for diagnosing breast cancer. The research aims to enhance breast cancer classification and localization through image classification and segmentation techniques utilizing mammography, ultrasound, and histopathology images. Among various image classification and segmentation techniques, the study selects technology and loss functions optimized for medical imaging characteristics, along with proposing data augmentation methods. The research findings demonstrate that using filter-based techniques for data augmentation yields excellent performance in image classification using ResNet50. Additionally, for the segmentation of mammography and ultrasound images, the UNet architecture performs exceptionally well. Through the application of these techniques, the segmentation performance of mammography images improved by 33.3%, ultrasound image segmentation improved by 29.9%, and histopathology image classification accuracy increased by 22.8%. This research presents a contribution to deep learning-based medical image processing in the context of breast cancer diagnosis.