Patient-Level Modeling and Classification of ST-segment in 12-Lead Electrocardiogram for Automated Myocardial...

Area of Science:

• Cardiology
• Biomedical Engineering
• Machine Learning

Background:

• Myocardial ischemia diagnosis relies on detecting ST-segment changes (ST elevation and ST depression) in ECGs.
• Accurate and timely detection of these changes is crucial for patient outcomes.

Purpose of the Study:

• To develop and evaluate a patient-level ST-segment classification approach using 12-lead ECGs.
• To compare the performance of various ensemble machine learning classifiers for ST-segment analysis.

Main Methods:

• Modeled ST-segments using least-squares polynomial regression to derive STp segments.
• Extracted handcrafted features from STp segments and their amplitude envelopes.
• Evaluated seven ensemble machine learning classifiers, including Light Gradient Boosting (LGB), for classification.

Main Results:

• The LGB classifier achieved the highest performance, with F1 scores of 0.94 (Normal), 0.86 (STD), and 0.72 (STE), averaging 0.84 for 12-lead ECGs.
• A six-lead ECG combination demonstrated an average F1 score of 0.83.
• Validation on the European ST-T database yielded high intra-patient F1 scores (0.87-1).

Conclusions:

• The proposed approach offers a clinically interpretable and transparent framework for ST-segment classification.
• This method shows state-of-the-art performance in detecting ST elevation and depression, supporting cardiologists in diagnosis.