Wangyixinxin / MRG-KG

Rethinking Medical Report Generation: Disease Revealing Enhancement with Knowledge Graph

MRG-KG

This repo holds code for Rethinking Medical Report Generation: Disease Revealing Enhancement with Knowledge Graph. (ICML IMLH 2023 Accepted)

Knowledge graph

KG.json

You can download our Disease Knowledge Graph in JSON format.

Disease augmentation

Aug_and_Eval.ipynb

You can download the IU-Xray any other datasets to perform a disease augmentation.

To augment your dataset, please refer to augmentation_to_count_sent() function.

Some analysis in our paper can be found, such as finding sentences w or w/o diseases, building disease pool/normal pool, and counting disease occurrences.

DOR and DS evaluation scores

Aug_and_Eval.ipynb

The proposed DS and DOR scores can be found in the Evaluation section.

Codes to generate Figure 1 (disease and sentence statistics) are also included.

Note: The generation model used in our paper is from R2Gen. Feel free to train it on the augmented dataset and evaluate the results using new scores.

**Update: We updated the implementation of DOR to avoid divide-by-zero issue.

Citation

If you find this paper, knowledge graph, or code useful for your research, please cite our paper:

@article{wang2023rethinking,
  title={Rethinking Medical Report Generation: Disease Revealing Enhancement with Knowledge Graph},
  author={Wang, Yixin and Lin, Zihao and Dong, Haoyu},
  journal={arXiv preprint arXiv:2307.12526},
  year={2023}
}