PerC-SimAM-BLSTM: Position Perception Circular Convolution With Simple Attention Mechanism Based on BLSTM for Bundle...

Area of Science:

• Artificial Intelligence in Medicine
• Cardiovascular Signal Processing
• Machine Learning for Healthcare

Background:

• Left bundle branch block (LBBB) and right bundle branch block (RBBB) are common arrhythmias with similar ECG features, complicating diagnosis.
• Conventional visual ECG inspection is challenging due to overlapping characteristics of LBBB and RBBB.
• Existing automated detection methods require improvement in accuracy and efficiency.

Purpose of the Study:

• To develop a novel MetaFormer-like model for accurate and efficient automatic detection of LBBB and RBBB.
• To enhance global-local feature acquisition for improved ECG analysis.
• To validate the model's performance against established arrhythmia databases.

Main Methods:

• A novel PerC-SimAM-BLSTM model was developed, combining position perception circular convolution (PerC) and a simple attention mechanism with bidirectional long short-term memory (SimAM-BLSTM).
• The model integrates Transformer's global feature extraction with PerC's enhanced global-local feature acquisition.
• Mathematical convergence proofs were used to ensure stable model training.

Main Results:

• The PerC-SimAM-BLSTM model achieved over 99.1% accuracy on the China Physiological Signal Challenge 2018 and MIT-BIH arrhythmia databases.
• The model demonstrated superior performance compared to existing strategies.
• The framework balances high performance with computational efficiency, suitable for lightweight devices.

Conclusions:

• The PerC-SimAM-BLSTM model offers a highly accurate and efficient solution for automated LBBB and RBBB detection from ECGs.
• This AI-driven approach shows significant potential for improving arrhythmia diagnosis, particularly in resource-constrained environments.
• The model's robust performance and efficiency make it a promising tool for clinical application and wearable health devices.