bonfab/chest-xray-nn-analysis

Investigating chest X-rays by: 1) training a neural network classifier and analyzing its predictions with LRP 2) Encoding images with auto-encoders to then detect outliers and patterns via clustering

ML in Medical Image Processing of Chest Xrays

Goals

Explainability

• Classification of X-Ray images
• Explainability of results with LRP

Outlier Detection

• Encode images with auto-encoders
• K-means clustering to detect outliers that often represent faulty images

Requirements

pip3 install --user --requirement requirements.txt

Alternatively, use one of the Docker images defined in the docker directory.

Usage

We used custom datasets in our Jupyter notebooks, these are available upon request.

Related

• LRP - Imported repo moboehle/Pytorch-LRP with own modifications
• docs/repo-tree.txt - Repo structure used to compute results in source code
• docs/links.md - Links to papers and datasets