Deep learning augmented ECG analysis to identify biomarker-defined myocardial injury

Area of Science:

• Artificial Intelligence in Medicine
• Cardiology
• Biomedical Engineering

Background:

• Chest pain is a frequent clinical issue where myocardial injury is a major concern, leading to significant morbidity and mortality.
• Early and accurate diagnosis of myocardial injury is crucial for effective patient management and improved outcomes.

Purpose of the Study:

• To develop and validate a deep convolutional neural network (CNN) capable of predicting serum troponin I (TnI) levels directly from electrocardiograms (ECGs).
• To assess the CNN's performance in classifying patients with and without elevated TnI, using different thresholds and ECG lead configurations.

Main Methods:

• A CNN model was trained using 64,728 ECGs from 32,479 patients with paired serum TnI results.
• The model predicted TnI levels (≥0.02 µg/L and ≥1.0 µg/L) and performed multi-class predictions.
• Performance was evaluated using Area Under the Curve (AUC) and tested on a separate cohort undergoing coronary angiography.

Main Results:

• The CNN accurately predicted elevated TnI at thresholds of 0.02 µg/L (AUC=0.783) and 1.0 µg/L (AUC=0.802).
• Models using single-lead ECGs showed slightly lower accuracy (AUCs 0.740-0.773).
• The model demonstrated similar performance in the coronary angiography cohort.

Conclusions:

• Deep convolutional neural networks can effectively predict biomarker-defined myocardial injury from both 12-lead and single-lead ECGs.
• This AI-driven approach shows promise for enhancing clinical decision-making in patients presenting with chest pain.