FlexCare

Source code for FlexCare: Leveraging Cross-Task Synergy for Flexible Multimodal Healthcare Prediction published in KDD 2024.

📄Paper is available at: ACM DL or arXiv

Requirements

This project is run in a conda virtual environment on Ubuntu 20.04 with CUDA 11.1.

• torch==1.10.1+cu111
• Python==3.7.9
• transformers==4.30.2
• tokenizers==0.13.3
• huggingface-hub==0.16.4

Data preparation

You will first need to request access for MIMIC dataset:

• MIMIC-IV 2.0 https://physionet.org/content/mimiciv/2.0/
• MIMIC-CXR-JPG 2.0.0 https://physionet.org/content/mimic-cxr-jpg/2.0.0/
• MIMIC-IV-NOTE 2.2 https://physionet.org/content/mimic-iv-note/2.2/

Then follow the steps in mimic4extract to build datasets for all tasks in directory [data].

In addition, we use biobert-base-cased-v1.2 as the pretrained text encoder, please download files in https://huggingface.co/dmis-lab/biobert-base-cased-v1.2, and put them into the directory [mymodel/pretrained]

Model training

python main_mt.py --data_path data --ehr_path data/ehr --cxr_path data/cxr --task in-hospital-mortality,length-of-stay,decompensation,phenotyping,readmission,diagnosis --epochs 25 --lr 0.0001 --device {gpu id} --seed {40,42,44,46,48}

Citation

@inproceedings{xu2024flexcare,
  title={FlexCare: Leveraging Cross-Task Synergy for Flexible Multimodal Healthcare Prediction},
  author={Xu, Muhao and Zhu, Zhenfeng and Li, Youru and Zheng, Shuai and Zhao, Yawei and He, Kunlun and Zhao, Yao},
  booktitle={Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining},
  pages={3610--3620},
  year={2024}
}