MAK-Net: A Multi-Scale Attentive Kolmogorov-Arnold Network with BiGRU for Imbalanced ECG Arrhythmia Classification

Area of Science:

• Cardiology
• Artificial Intelligence
• Biomedical Signal Processing

Background:

• Accurate electrocardiogram (ECG) classification is crucial for diagnosing arrhythmias.
• Real-world ECG datasets often exhibit severe class imbalance, hindering diagnostic performance.
• Existing methods struggle with imbalanced data, impacting recall and F1-scores.

Purpose of the Study:

• To introduce MAK-Net, a hybrid deep learning framework designed to overcome class imbalance in ECG signal classification.
• To enhance the accuracy and robustness of automated arrhythmia detection.

Main Methods:

• Developed MAK-Net, a hybrid deep learning framework integrating multiscale convolutional modules, channel attention, bidirectional gated recurrent units (BiGRU), and Kolmogorov-Arnold Network (KAN) layers.
• Employed focal loss and Synthetic Minority Oversampling Technique (SMOTE) to address data imbalance.
• Utilized the MIT-BIH arrhythmia database for model evaluation.

Main Results:

• MAK-Net achieved state-of-the-art performance on the MIT-BIH arrhythmia database.
• Achieved high metrics: 0.9980 accuracy, 0.9888 F1-score, 0.9871 recall, 0.9905 precision, and 0.9991 specificity.
• Demonstrated superior robustness to imbalanced classes compared to existing methods.

Conclusions:

• The proposed MAK-Net framework effectively handles imbalanced ECG data for reliable arrhythmia detection.
• Multiscale feature fusion, attention-guided learning, and KAN-based nonlinear mapping are validated for automated arrhythmia diagnosis.
• MAK-Net offers a promising solution for clinically reliable automated arrhythmia detection.