MLVICX: Multi-Level Variance-Covariance Exploration for Chest X-ray Self-Supervised Representation Learning

The code repository of IEEE JBHI 2024 paper MLVICX: Multi-Level Variance-Covariance Exploration for Chest X-ray Self-Supervised Representation Learning

Structure of this repository

This repository is organized as:

• config YAML files
• data preprocessed data and dataloaders
• models MLVICX model architecture
• optimizers LARS for SGD
• trainer trainer for MLVICX
• utils utility functions
• main.py training MLVICX

Usage Guide

Dataset Preparation

NIH-CXR 14 Dataset

NIH data is available here

We resize images to 224 x 224 for training. Preprocessed CSV files are provided in /data/. Please use the provided dataloaders in /data/ to load data.

Please don't forget to update paths in /config/mlvicx.yaml

The process is similar to training on any other dataset.

Running

Training MLVICX

Once the dataset is set up, update the necessary hyperparameters according to the paper to reproduce any experiments and run,

python main.py --mode ssl --init rand --bs 64 --epoch 300 --dataset nih --seed 42 --gpu 0 --resume False 

After training, the checkpoints will be stored in /ckpt as assigned. Check the trainer for making any changes.

If you want to try different models, use --model. For fine-tuning, change mode --mode sl.

Citation

If you find our work useful please cite as,

@article{mlvicx,
  author={Singh, Azad and Gorade, Vandan and Mishra, Deepak},
  journal={IEEE Journal of Biomedical and Health Informatics}, 
  title={MLVICX: Multi-Level Variance-Covariance Exploration for Chest X-Ray Self-Supervised Representation Learning}, 
  year={2024},
  volume={28},
  number={12},
  pages={7480-7490}}