Lead importance in 12-lead ECG
Project proposed for graduate level class: Machine Learning for Medical Application ML4MA_project.pdf
This project investigates the relative importance of each lead in a 12-lead electrocardiogram (ECG) for diagnosing cardiac conditions. By optimizing lead selection, we aim to improve the efficiency and accessibility of ECG diagnostics, particularly in resource-constrained environments.
- Train models to predict cardiac conditions using a reduced number of ECG leads.
- Evaluate the diagnostic value of each lead using Shapley values.
- Validate models on external datasets to ensure generalizability.
Cardiovascular diseases (CVDs) are a leading cause of mortality worldwide. While 12-lead ECGs are valuable diagnostic tools, their complexity may limit accessibility. Reducing the number of required leads can:
- Lower costs
- Improve accessibility in underserved areas
- Enhance patient comfort
- Expedite diagnosis without compromising accuracy
- Cost-Effective Diagnosis: Streamlines workflows and reduces financial burden.
- Time Efficiency: Accelerates clinical decision-making.
- Accessibility: Enables diagnostics in remote or resource-poor settings.
- Personalized Care: Facilitates tailored diagnostic approaches.
- Advancing Knowledge: Contributes to ECG analysis methodologies and clinical best practices.
- Monte Carlo sampling approach to estimate the contribution of each lead.
- Enables efficient evaluation without exhaustive computation of all lead combinations.
Download A large scale 12-lead electrocardiogram database for arrhythmia study from physionet.
Downloaded from here.
Run data_prep.ipynb and data_prep_ptb-xl.ipynb.
Run train_model_mult_outputs.ipynb. For single load models run single_lead_mult_outputs.ipynb.
For coalitions run train_6_lead_model.ipynb.
Arrythmia dataset: evaluate_internal_test.ipynb.
PTB-XL: eval_ptb-xl.ipynb.
6 lead models: eval_6_lead_model.ipynb.
Run shap_approximation_all.ipynb.
- Zheng, J., et al. A large scale 12-lead electrocardiogram database for arrhythmia study. PhysioNet (2022).
- Wagner, P., et al. PTB-XL: A large publicly available electrocardiography dataset. Sci. Data (2020).
- Attia, Z. I., et al. Screening for cardiac contractile dysfunction using AI-enabled ECG. Nat. Medicine (2019).
- Štrumbelj, E., et al. Explaining prediction models and individual predictions with feature contributions. Knowl. Inf. Sys. (2014).