References & Citations
Electrical Engineering and Systems Science > Image and Video Processing
Title: Improving Mammography Malignancy Segmentation by Designing the Training Process
(Submitted on 29 May 2020)
Abstract: We work on the breast imaging malignancy segmentation task while focusing on the training process instead of network complexity. We designed a training process based on a modified U-Net, increasing the overall segmentation performances by using both, benign and malignant data for training. Our approach makes use of only a small amount of annotated data and relies on transfer learning from a self-supervised reconstruction task, and favors explainability.
Link back to: arXiv, form interface, contact.