HPRNet: a hierarchical pyramidal residual network for ECG arrhythmia classification.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence
• Cardiology

Background:

• Automated cardiac arrhythmia diagnosis relies heavily on accurate electrocardiogram (ECG) signal classification.
• Non-stationary signals and noise in ECG recordings pose significant challenges for existing deep learning models, hindering robust feature extraction.

Purpose of the Study:

• To propose a novel deep learning model, the Hierarchical Pyramidal Residual Network (HPRNet), for improved ECG arrhythmia classification.
• To enhance feature extraction capabilities for noisy and non-stationary ECG signals.
• To optimize model efficiency through parameter reduction.

Main Methods:

• Developed HPRNet featuring a Hierarchical Pyramidal REB-based Backbone (HRB) to capture multi-scale ECG signal characteristics.
• Implemented a Multi-Level Pruning Optimization (MLPO) strategy for parameter reduction and computational efficiency.
• Evaluated HPRNet on the MIT-BIH and INCART public benchmark datasets.

Main Results:

• HPRNet achieved superior performance compared to five representative methods on the MIT-BIH dataset, reaching an F1-score of 92.05%.
• On the INCART dataset, HPRNet obtained a 91.98% accuracy for binary classification with an average inference latency of 0.031 seconds.
• Ablation studies confirmed the effectiveness of the proposed HRB and MLPO strategies.

Conclusions:

• HPRNet demonstrates robustness and superior performance in automated ECG arrhythmia classification, addressing challenges posed by signal non-stationarity and noise.
• The proposed model offers an efficient and accurate solution for clinical applications in cardiac arrhythmia diagnosis.
• The findings support the potential of HPRNet for reliable automated diagnosis of cardiac arrhythmias.