2

SegWithDistMap: Distance Transform Maps for Medical Image Segmentation

An empirical study on how Distance Transform Maps boost segmentation CNNs in medical imaging tasks.

SegWithDistMap is a comprehensive project that explores various methods of incorporating distance transform maps into CNN-based medical image segmentation tasks. This repository implements and evaluates multiple approaches published in 2019, focusing on 3D segmentation tasks for heart and liver tumor segmentation.

Key features:

  • Implementation of various distance transform map methods:
    • New loss functions (e.g., Boundary loss, Hausdorff loss)
    • Auxiliary tasks (e.g., distance map regression)
  • Evaluation on 3D medical image segmentation tasks
  • Comparison of different approaches on the same datasets
  • Extensive experiments with various GPU setups

This project provides valuable insights into the effectiveness of distance transform maps in improving segmentation performance across different medical imaging tasks.

View on GitHub

Related Paper