Compact neural network algorithm for electrocardiogram classification.

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence in Medicine

Background:

• Current electrocardiogram (ECG) analysis heavily relies on deep learning (DL) and convolutional neural networks (CNNs).
• DL models demand significant computational resources and large labeled datasets, limiting their application in resource-constrained environments.
• There is a need for efficient and interpretable ECG classification algorithms suitable for widespread clinical use.

Purpose of the Study:

• To develop a powerful, compact ECG classification algorithm for cardiac arrhythmia diagnosis.
• To reduce the dependency on computationally intensive deep learning models in ECG analysis.
• To create an interpretable model deployable on resource-limited platforms.

Main Methods:

• Developed an artificial neural network (ANN) with a simple architecture.
• Implemented a compact, interpretable feature-engineering pipeline using 17 engineered features.
• Integrated mathematical transformations, signal processing, and data extraction for efficient ECG pattern recognition.
• Avoided deep learning and convolutional neural networks (CNNs).

Main Results:

• The algorithm achieved performance comparable to state-of-the-art models in classifying five key cardiac rhythms.
• Achieved high accuracy ([Formula: see text]), Cohen's kappa ([Formula: see text]), and Matthews correlation coefficient ([Formula: see text]) on MIT-BIH and St. Petersburg INCART databases.
• Demonstrated low inference latency on a standard CPU, highlighting its efficiency.

Conclusions:

• The proposed ANN-based ECG classification algorithm offers a viable alternative to deep learning approaches.
• The compact model and efficient feature engineering enable deployment on resource-constrained platforms for cardiac arrhythmia diagnosis.
• This approach facilitates broader accessibility and application of advanced ECG analysis in clinical settings.