amartyacodes / Multi-modal-Fusion-Model-for-Predicting-Adverse-Cardiovascular-Outcomes

Multi-modal-Fusion-Model-for-Predicting-Adverse-Cardiovascular-Outcomes

The code for the multi-modal fusion model for prediction of adverse cardiovascular outcomes using both ECG and EHR Data. The code was implemented using and

Manuscript

https://iopscience.iop.org/article/10.1088/1361-6579/ac9e8a

Model Description

The model has been built for prediction of cardiovascular outcomes such as Stroke, Chronic Heart Failure and Mortality using the ECG data as well as the Electronic Health Record(EHR) data.

The ECG Data was extracted from the images and preprocessed before being sent to the model

ECG Image

Preprocessed ECG Image

Model

Files Present

1. utility.py - Contain the metrics
2. train.py - The code for building the required model and perform the training operation
3. requirements.txt - The libraries required to run the code
4. plotsave.py - Plots the ROC-AUC curve
5. main.py - Running the code and saving the model results
6. ROC.png - Sample ROC curve generated
7. dataread.py- Reading the ECG data, the EHR data, and the labels
8. config.json- Used in building the branch corresponding to the ECG data and controlling the number of convolution layers
9. ModelImage.png - Picture representation of our proposed fusion model

How to Implement?

Initially create an environment and install the requirements using the code below.

pip install -r /path/to/requirements.txt

Then run the main.py file, giving which takes the input a CSV file containing EHR data, using the code present below. The EHR data was named as "Share_data_outcome_deidentified.csv" in our case

!python3 main.py -i Share_data_outcome_deidentified.csv

This will automatically save the ROC.png file having the ROC scores The data we worked on has not been uploaded yet.

Result

The results obtained and the comparisons have been mentioned in our manuscript.

Contribution

The code contribution of the Fusion Model was made Amartya Bhattacharya while working as a research assistant at Banerjee Lab, Arizona State University under the guidance of Dr. Imon Banerjee and Dr. Arindam Sanyal.

I wish to convey my sincere gratitude to the professors for giving me the opportunity to work on this project.