Automated Breast Tumor Segmentation in DCE-MRI Using Deep Learning

Segmentation of breast tumor is an important step for breast cancer follow-up and treatment. Automating this challenging task can help radiologists to reduce the high workload of breast cancer analysis. In this paper, we propose a deep learning approach to automate the segmentation of breast tumors in DCE-MRI data. We build an architecture based on U-net fully convolutional neural network. The trained model can handle both detection and segmentation on each single breast slice. In this study, we used a dataset of 86 DCE-MRI, acquired before and after chemotherapy, of 43 patients with local breast cancer, a total of 5452 slices. The data have been annotated manually by an experienced radiologist. The model was trained and validated on 85% and 15% of the data and achieved a mean IoU of 76,14%.