Interpretable deep learning for automatic diagnosis of 12-lead electrocardiogram
Problem
Growing ECG exam volumes and a shortage of cardiologists make automatic, accurate arrhythmia diagnosis from 12-lead ECGs necessary.
Why it matters
Cardiovascular disease diagnosis at scale depends on relieving the cardiologist bottleneck.
What the AI did
A deep neural network classified cardiac arrhythmias from 12-lead ECG recordings, outperforming four expert-feature machine learning methods, with SHAP used for patient- and population-level interpretation.
Results
Average F1 score of 0.813 on a public 12-lead ECG dataset; superior to 4 machine learning baselines
Impact
Cited 189 times; demonstrated interpretable automated ECG classification on a public dataset.
Record
| Venue | iScience |
|---|---|
| Published | 2021 |
| Work type | article |
| Authors | Dongdong Zhang, Samuel Yang, Xiaohui Yuan, Ping Zhang |
| Citations | 189 (OpenAlex, at import time) |
| References | 50 |
| Model or method | deep neural network (unnamed), interpreted with SHAP |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1016/j.isci.2021.102373 |
| Open access | gold |
| Keywords | ECG, deep learning, arrhythmia, interpretable AI, SHAP, cardiology |
Importance rating
significant — thousands to millions affected
Abstract cites rapid growth of ECG exams and cardiologist insufficiency; automated arrhythmia diagnosis addresses cardiovascular disease diagnosis at scale, but no quantified impact is given.