denizlab/OAI-KL-Grade-Classification

code for paper Attention-based CNN for KL Grade Classification: Data from the Osteoarthritis Initiative

About

This Repo contains code for paper Attention-based CNN for KL Grade Classification: Data from the Osteoarthritis Initiative

Repo structure

• ./data contains data for training/testing data for detector and classifier. OAI_summary.csv file is from OAI dataset, and contains metadata for all patients. The train/test split gave the performance mentioned in paper.
• ./model_weights contains model weights that can be readily used by torch.load with the performance mentioned in paper.
• ./oai-knee-detection contains code to train a detector and generate all annotations
• ./oai-xray-klg contains code to train classifier and generate the attention map from GradCAM

Instruction

Please refer to requirements.txt for install all dependencies for this project. ./data folder contains example content file for train/test data used in dataloader for both detector and classifier. ./model_weights folder contains model weights that achieved the performance metrics mentioned in paper.

This repo consists of two parts. To reproduce the entire experiments, you will need to

1. Train a detector and use the detector to annotate all OAI dataset, and generate train/test data for the classifier. See documentation in ./oai-knee-detection.
2. Train and test the classifier by following documentation in ./oai-xray-klg

How to cite

@inproceedings{zhang2020attention,
  title={Attention-based cnn for kl grade classification: Data from the osteoarthritis initiative},
  author={Zhang, Bofei and Tan, Jimin and Cho, Kyunghyun and Chang, Gregory and Deniz, Cem M},
  booktitle={2020 IEEE 17th International Symposium on Biomedical Imaging (ISBI)},
  pages={731--735},
  year={2020},
  organization={IEEE}
}