This project aims to perform a bias audit on the FCT model for different BMI and Age groups on the ACDC dataset.

Automated Cardiac Diagnosis Challenge Dataset available here.

Visualizations, analysis were performed to get a better understanding of the dataset.

The Fully Convolutional Transformer (FCT) proposed by this paper is used in this project. The model is trained on the ACDC dataset using the PyTorch code variant available here.

Trained the model for 120 epochs which took around 3 hours on a NVIDIA Tesla T4 GPU. Then, the model was saved for further analysis and auditing. Model is available here.

The FCT model was evaluated on the test set of 50 patients. Analysis will follow soon.

1. Get ACDC dataset and delete .md file in your ACDC dataset folder
2. use python main.py to start training