Improving Mammography Malignancy Segmentation by Designing the Training Process

TL;DR

This paper enhances mammography malignancy segmentation by optimizing the training process with transfer learning and combined benign and malignant data, achieving better performance with limited annotations.

Contribution

It introduces a training methodology that leverages self-supervised transfer learning and combined data to improve segmentation accuracy without increasing network complexity.

Findings

Improved segmentation performance using the proposed training process.

Effective use of limited annotated data through transfer learning.

Enhanced explainability of the segmentation model.

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.