COVID-19 Detection from Chest Radiography Images

This work compares three deep learning approaches for COVID-19 detection from chest radiographies. Our first method performs transfer learning with a pre-trained ResNet-50. Further, we propose a model that relies on anomaly detection with the U-Net architecture and a ResNet-50 classifier. Our third approach is based on multitask learning where a modified U-Net performs a reconstruction task and a classification task simultaneously.

Setup Instructions

Create virtual environment and follow these steps.

git clone [email protected]:axkoenig/dl4mi.git
pip install -e dl4mi

Download the COVIDx3 dataset presented in the COVID-Net paper. Please follow instructions in the official repository to download the dataset. Place the data in a directory of your choice. The subdirectories of your data directory should be called "train" and "test". Place the respective files in these directories.

The models can be trained using the classifier.pyscript in the respective directories. To check what arguments can be specified run the script with the -h flag.

python 3_multitask_learning/classifier.py -h