/PM2F2N

The source code for Findings of EMNLP 2022 paper "PM2F2N: Patient Multi-view Multi-modal Feature Fusion Networks for Clinical Outcome Prediction"

Primary LanguagePython

PM2F2N

[Title] PM2F2N: Patient Multi-view Multi-modal Feature Fusion Networks for Clinical Outcome Prediction

[Authors] Ying Zhang, Baohang Zhou, Kehui Song, Xuhui Sui, Guoqing Zhao, Ning Jiang and Xiaojie Yuan

[EMNLP 2022 Findings]

Preparation

  1. Clone the repo to your local.
  2. Download Python version: 3.6.13
  3. Download the preprocessed data from this link and the extraction code is 1234. Put the downloaded files into the data folder.
  4. Open the shell or cmd in this repo folder. Run this command to install necessary packages.
pip install -r requirements.txt

Experiments

  1. For Linux systems, we have shell scripts to run the training procedures. You can run the following command:
./train.model.sh
  1. You can also input the following command to train the model. There are different choices for some hyper-parameters shown in square barckets. The meaning of these parameters are shown in the following tables.
Parameters Value Description
epoch int Training times
patience int Early stopping
weights string Saved model path
save_features int Whether to save multimodal features
task string Choose the clinical outcome task 
CUDA_VISIBLE_DEVICES=0 python mmg_main.py \
    --seed 0 \
    --epoch 300 \
    --patience 20 \
    --weights ./weights/mort_icu.h5 \
    --task mort_icu \
    --mode train \
    --save_features 0
  1. After training the model, you can run the test script to evaluate the model on the test set.
./test.model.sh

or

CUDA_VISIBLE_DEVICES=0 python mmg_main.py \
    --seed 0 \
    --weights ./weights/mort_icu.h5 \
    --task mort_icu \
    --mode test \
    --save_features 0
  1. We also provide the weights of the model to reimplement the results in our paper. The saved weights have been in weights folder. You can run the test script directly after downloading the processed data.

References

  1. We utilize the MIMIC-Extract pipline tools to acquire the raw data. And you can also follow the guide lines to process the raw data for your research.