This repository hosts the code of the paper Prospects for AI-Enhanced ECG as a Unified Screening Tool for Cardiac and Non-Cardiac Conditions -- An Explorative Study in Emergency Care. Accepted by European Heart Journal Digital Health. In this study we introduced a unified deep learning model for ECG analysis, predicting a wide range of cardiac and non-cardiac discharge diagnoses based on the ICD10 classification system with impressive AUROC scores. Our approach excels in handling diverse diagnostic scenarios, suggesting its use as a screening tool in emergency departments, integrated into clinical decision support systems. We therefore propose the MIMIC-IV-ECG-ICD-ED dataset derived from the MIMIC-IV and MIMIC-IV-ECG databases primarily for benchmark purposes.

(A) represents the distribution of statements according to chapters (all percentages as relative fractions compared to the dataset size), whereas (B) represents the distribution of cardiac conditions within chapter IX.

(A) represents the distribution of statements according to chapters (all percentages as relative fractions compared to the dataset size), whereas (B) represents the distribution of cardiac conditions within chapter IX. However, these are the distributions of a specific ED use case (subset dataset) investigated in the manuscript.

Download the MIMIC-IV-ECG dataset and the MIMIC-IV dataset (with credentialed access).

Go under src/ and run the following command where your should replace the corresponding data paths

python full_preprocessing.py --mimic-path <path to mimic-iv directory ended in 'mimiciv/2.2/'> --zip-path <path to ecgs zip file> --target-path <desired output for preprocessed data default='./'>

These are 2 of the total benchmarks commands, T(ED2ALL)-E(ED2ALL) the main scenario thtought the main text, and T(ALL2ALL)-E(ALL2ALL) the complete dataset. These command should also export your test set predictions into a corresponding path directory (already specified in a command argument), and also save resulting AUROCs in an also specified log file.

Optinally, see src/demo.ipynb for an example of how to acess each of specific bencharmking scenario.

T(ED2ALL)-E(ED2ALL)

python main_ecg.py --data <your data path> --input-size 250 --finetune-dataset mimic_ed_all_edfirst_all_2000_5A --architecture s4 --precision 32 --s4-n 8 --s4-h 512 --batch-size 32 --epochs 20 --export-predictions-path T(ED2ALL)-E(ED2ALL)/ > T(ED2ALL)-E(ED2ALL).log

T(ALL2ALL)-E(ALL2ALL)

python main_ecg.py --data <your data path> --input-size 250 --finetune-dataset mimic_all_all_allfirst_all_2000_5A --architecture s4 --precision 32 --s4-n 8 --s4-h 512 --batch-size 32 --epochs 20 --export-predictions-path T(ALL2ALL)-E(ALL2ALL)/ > T(ALL2ALL)-E(ALL2ALL).log

You can find all the experimental results for each of the labels and scenarios under reports/Results_MIMIC_IV_ECG_ICD.csv

@article{strodthoff2024prospects,
  title={Prospects for artificial intelligence-enhanced electrocardiogram as a unified screening tool for cardiac and non-cardiac conditions: an explorative study in emergency care},
  author={Strodthoff, Nils and Lopez Alcaraz, Juan Miguel and Haverkamp, Wilhelm},
  journal={European Heart Journal-Digital Health},
  pages={ztae039},
  year={2024},
  publisher={Oxford University Press UK}
}