MedAI: Transparency in Medical Image Segmentation

What is this repo

This repo contains the code and experiments that are implemented to contribute in MedAI Cahllenge: Transparency in Medical Image Segmentation to automate medical image segmentation while preserve transparency. In the paper below we proposed generative adversarial network-based models to segment both polyps and instruments in endoscopy images. We also provide explanations for the predictions using a layer-wise relevance propagation approach.

Reference papers

Working paper:

Explainable Medical Image Segmentation via Generative Adversarial Networks and Layer-wise Relevance Propagation, Awadelrahman Ahmed, Leen Ali, Nordic Machine Intelligence. https://journals.uio.no/NMI/article/view/9126

Challenge details paper:

BibTeX:
@article{MediAI2021, title = {{MedAI: Transparency in Medical Image Segmentation}}, author = { Hicks, Steven and Jha, Debesh and Thambawita, Vajira and Riegler, Michael and Halvorsen, P{\aa}l and Singstad, Bj{\o}rn-Jostein and Gaur, Sachin and Pettersen, Klas and Goodwin, Morten and Parasa, Sravanthi and de Lange, Thomas }, journal = {Nordic Machine Intelligence},
year = {2021}, doi = {10.5617/nmi.9140}

Model

Sample Results

Polyp secmentation task sample

Instrument secmentation task sample

Evaluation