Precision-guided gramian angular field for ECG-based cardiac arrhythmia detection utilizing binary-weight simplicial convolutional neural networks

Abstract

Cardiac arrhythmia, characterized by abnormal heart rhythms, can be effectively identified using electrocardiogram (ECG) signals. However, the complex temporal and frequency patterns in ECG recordings make accurate diagnosis challenging. Recent advances in deep learning have significantly improved automated arrhythmia detection. This paper proposes a Precision-Guided Gramian Angular Field–based ECG classification framework utilizing Binary-Weight Simplicial Convolutional Neural Networks (PGDWF-BWSCNN). ECG signals are obtained from a standard ECG Arrhythmia Classification Dataset and preprocessed using an Isolated Kalman Filter (IKF) to remove noise and artifacts such as baseline wander, power-line interference, and muscle noise. The denoised signals undergo feature extraction using the Fractional Shehu Transform (FST) to capture both temporal and frequency-domain characteristics. These features are then classified by the Binary-Weight Simplicial Convolutional Neural Network (BWSCNN) into five categories: Normal, Supraventricular ectopic beat, Ventricular ectopic beat, Fusion beat, and Unknown beat.To enhance convergence and overall performance, the Battlefield Optimization Algorithm (BOA) is employed to optimize the BWSCNN by dynamically adjusting learning parameters. The proposed model is evaluated using Accuracy, Sensitivity, F1-score, and Confusion Matrix. Experimental results demonstrate that PGDWF-BWSCNN achieves 99.46% accuracy and 98.63% sensitivity, outperforming existing approaches including ECGHC-RNN-DCNN, ADEC-ResNet, and MADECG-CNN. These results confirm the effectiveness of the proposed method for accurate and reliable multi-class arrhythmia classification.