Typicality excels Likelihood for Unsupervised Out-of-Distribution Detection in Medical Imaging

This repository represents the code of the paper Typicality Excels Likelihood for Unsupervised Out-of-Distribution Detection in Medical Imaging.

The code is adapted from here.

Data

The datasets used in this work can be accessed below:

Dataset name	Imaging modality	Download link
ISIC	Dermoscopic RGB images	Link
COVID-19	Chest X-Ray	Link
Pneumonia	Chest X-Ray	Link
HeadCT	Brain CT (2D)	Link

Dependencies

Setup the conda environment with environment.yml:

conda env create -f environment.yml

Active the typicalmood conda environment:

source activate
conda activate typicalmood

Training

Train the model on the ISIC dataset:

python train.py --data isic --batch_size 8 --alpha 2.0 --patience 10 

Other arguments can be seen by calling: python train.py -h.

The configuration of the GLOW model can be adjusted in config_glow.py. The hyperparameters used to train on ISIC can be adjusted in isic.py.