Explainable Medical Image Segmentation via Generative Adversarial Networks and Layer-wise Relevance Propagation

TL;DR

This paper introduces a GAN-based medical image segmentation method that provides explainability through layer-wise relevance propagation, achieving high accuracy and Jaccard scores on endoscopy images.

Contribution

It combines generative adversarial networks with explainability techniques to improve medical image segmentation and interpretability.

Findings

Achieved 0.84 accuracy and 0.46 Jaccard for polyp segmentation.

Achieved 0.96 accuracy and 0.70 Jaccard for instrument segmentation.

Provides explainability for model predictions using relevance propagation.

Abstract

This paper contributes to automating medical image segmentation by proposing generative adversarial network-based models to segment both polyps and instruments in endoscopy images. A major contribution of this work is to provide explanations for the predictions using a layer-wise relevance propagation approach designating which input image pixels are relevant to the predictions and to what extent. On the polyp segmentation task, the models achieved 0.84 of accuracy and 0.46 on Jaccard index. On the instrument segmentation task, the models achieved 0.96 of accuracy and 0.70 on Jaccard index. The code is available at https://github.com/Awadelrahman/MedAI.