ECG Recurrence Plot-Based Arrhythmia Classification Using Two-Dimensional Deep Residual CNN Features.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Signal Processing

Background:

• Arrhythmia classification is crucial for cardiac health management.
• Existing algorithms often face limitations in computational resources for portable applications.
• Recurrence plots (RPs) offer a novel approach for time-series analysis in ECG data.

Purpose of the Study:

• To develop an effective and memory-efficient electrocardiogram (ECG) recurrence plot (RP)-based arrhythmia classification algorithm for portable devices.
• To improve classification accuracy through a two-stage deep learning approach.
• To evaluate the algorithm's performance against established benchmarks and previous studies.

Main Methods:

• ECG time series data from PhysioNet databases were segmented and converted into recurrence plots (RPs).
• A two-stage Convolutional Neural Network (CNN) classification strategy was employed, utilizing ResNet-18 for initial detection and ResNet-50 for refined classification.
• The model was validated using 5-fold cross-validation.

Main Results:

• The algorithm achieved high average accuracies of 97.21% (Stage 1) and 98.36% (Stage 2).
• Sensitivities reached 96.49% and 97.92%, with positive predictive values of 95.54% and 98.20%, respectively.
• A significant 5-fold reduction in memory requirements was observed, enhancing feasibility for portable devices.

Conclusions:

• The proposed ECG RP-based arrhythmia classification algorithm is effective and suitable for resource-constrained portable devices.
• The two-stage ResNet architecture demonstrates superior performance in accuracy and efficiency.
• Data imbalance in the first stage, particularly with the 'other' category, requires further investigation to optimize performance.