AI-Based Breast Cancer Diagnosis Using Ultrasound and MRI Images Through Deep Learning

Developed a deep learning-based medical imaging project focused on breast cancer lesion detection and segmentation using ultrasound and MRI images. Implemented and evaluated multiple deep learning architectures, including U-Net++, U-Net3+, Attention U-Net, TransUNet, DeepLabV3, and YOLOv11. Conducted experiments using the BUS-UCLM and Duke Breast Cancer MRI (MAMA-MIA) datasets. Designed two complementary approaches: direct lesion segmentation and a two-stage detection-and-segmentation pipeline. Evaluated model performance using Dice Score, IoU, precision, recall, and mAP. Achieved Dice Scores exceeding 94% for ultrasound and 96% for MRI in the two-stage approach.