Scholarlink

<p>Arrhythmia detection is crucial in preventive cardiology, and deep learning (DL) enhances this by learning complex patterns from ECG signals. Ventricular Arrhythmias, ranging from isolated premature beats to sustained tachyarrhythmia’s, increase the risk of sudden cardiac death. The study presents a hybrid pipeline that merges Convolutional Neural Network-based deep feature extraction with SVM classification to enhance detection accuracy and generalization. ECG-FlexPrep is a pre-processing engine that normalizes ECG data, while the PulseNet-1D 1D CNN extracts detailed patterns to create deep embeddings. These embeddings are classified using the CardioBeat Vector Class (CVBC) Net, and the RhythmBlend ensemble strategy enhances classification by integrating outputs from both CNN and SVM. This framework uses GradCam, heat maps, and ROC curves to evaluate performance. These modules combine to provide an accurate, interpretable, and generalizable arrhythmia classification system that outperforms CNNs and standalone SVM classifiers.The proposed deep hybrid rhythm blend ensemble models achieved superior outcomes in ventricular arrhythmia classification. As shown in Table&#xa0;<InternalRef RefID="Tab1">1</InternalRef>, the CardioBeat Vector Class Net (CBVC-Net) recorded recall, precision, and F1-score values of 97.72%, outperforming RhythmBlend soft voting at 96.99% and hard voting at about 91%. In contrast, the existing ResNet34 model lagged with a precision of 80.34% and F1-score of 80.61%. Table <InternalRef RefID="Tab2">2</InternalRef> further highlights CBVC-Net’s accuracy across ventricular tachycardia (97.31%), ventricular fibrillation (98.06%), and non-ventricular rhythms (97.79%), consistently surpassing baselines.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Analytical Study of Ventricular Arrhythmia Classification Using Deep Hybrid Rhythm Blend Ensemble Algorithms

  • 期刊论文

  • 发表日期：

    2026年3月26日

查看全文
  • M. S. Supriya,
  • K. S. Arvind,
  • Manikandan Parasuraman

摘要

Arrhythmia detection is crucial in preventive cardiology, and deep learning (DL) enhances this by learning complex patterns from ECG signals. Ventricular Arrhythmias, ranging from isolated premature beats to sustained tachyarrhythmia’s, increase the risk of sudden cardiac death. The study presents a hybrid pipeline that merges Convolutional Neural Network-based deep feature extraction with SVM classification to enhance detection accuracy and generalization. ECG-FlexPrep is a pre-processing engine that normalizes ECG data, while the PulseNet-1D 1D CNN extracts detailed patterns to create deep embeddings. These embeddings are classified using the CardioBeat Vector Class (CVBC) Net, and the RhythmBlend ensemble strategy enhances classification by integrating outputs from both CNN and SVM. This framework uses GradCam, heat maps, and ROC curves to evaluate performance. These modules combine to provide an accurate, interpretable, and generalizable arrhythmia classification system that outperforms CNNs and standalone SVM classifiers.The proposed deep hybrid rhythm blend ensemble models achieved superior outcomes in ventricular arrhythmia classification. As shown in Table 1, the CardioBeat Vector Class Net (CBVC-Net) recorded recall, precision, and F1-score values of 97.72%, outperforming RhythmBlend soft voting at 96.99% and hard voting at about 91%. In contrast, the existing ResNet34 model lagged with a precision of 80.34% and F1-score of 80.61%. Table 2 further highlights CBVC-Net’s accuracy across ventricular tachycardia (97.31%), ventricular fibrillation (98.06%), and non-ventricular rhythms (97.79%), consistently surpassing baselines.