Automated arrhythmia classification based on a pyramid dense connectivity layer and BiLSTM.

Area of Science:

• Cardiology
• Artificial Intelligence
• Signal Processing

Background:

• Deep neural networks (DNNs) are increasingly used for automatic arrhythmia classification.
• Current DNN models show limitations in classification accuracy.
• There is a need for improved methods to capture detailed temporal and inter-channel information.

Purpose of the Study:

• To develop a more effective approach for automatic arrhythmia classification.
• To enhance receptive field sizes for capturing multi-scale temporal information.
• To incorporate inter-channel correlations for improved feature extraction.

Main Methods:

• Proposed a pyramidal dense connectivity layer and bidirectional long short-term memory network (PDC-BiLSTM).
• Integrated efficient channel attention (ECA) for dynamic feature channel weighting.
• Evaluated the model on the MIT-BIH arrhythmia database.

Main Results:

• Achieved 99.82% overall accuracy in the intra-patient paradigm.
• Reported 99.64% positive predictive value, 97.61% sensitivity, and 98.60% F1 Score (intra-patient).
• Attained 96.30% overall accuracy in the inter-patient paradigm.

Conclusions:

• The proposed PDC-BiLSTM with ECA model outperforms existing methods in arrhythmia classification accuracy.
• Demonstrates significant potential for application in cardiovascular disease diagnostic devices.
• Highlights the effectiveness of combining multi-scale feature extraction with attention mechanisms.