[Paper] Integration of Multi-Source Medical Data for Medical Diagnosis Question Answering

The repository is for the paper "Integration of Multi-Source Medical Data for Medical Diagnosis Question Answering"

The main contributions can be summarized as follows: (i) We explore a medical question-answering task based on multi-source data, named medical diagnosis question answering (MedDQA) task. (ii) We introduce a new dataset constructed from real medical data as a foundation for future research. (iii) We propose a multi-agent system where each agent can selectively handle specific data sources.

Requirements

Install all required python dependencies:

pip install -r requirements.txt

Dataset

Please check the Google Drive: https://drive.google.com/file/d/1zHmIb7Ej3lb3z6KO_pT9IfX0YBZJapFF/view?usp=sharing

Note that Our dataset is currently undergoing ethical review by the hospital. It will be made the whole dataset publicly available once the review process is completed.

Implementation

1. Place the Dataset

Place the downloaded dataset in the ./MMA_Dataset directory, including the json_file folder and image folder.

2. Configure the API Key

Add your API key to the prompt_generator.py file.

3. Run the Python File

Run the file using the following command:

python run.py

Citation

If this repo is useful to you, please cite using this BibTeX.

@article{peng2024integration,
  title={Integration of Multi-Source Medical Data for Medical Diagnosis Question Answering},
  author={Peng, Qi and Cai, Yi and Liu, Jiankun and Zou, Quan and Chen, Xing and Zhong, Zheng and Wang, Zefeng and Xie, Jiayuan and Li, Qing},
  journal={IEEE Transactions on Medical Imaging},
  year={2024},
  publisher={IEEE}
}