Scholarlink

<p>Ventricular Fibrillation (VF), a deadly cardiac arrhythmia, kills many unexpectedly, making early prediction crucial. The discovery plugs a gap in bespoke prediction systems that lack generalizability across patient profiles and real-time interpretability. To overcome these limitations, a novel hybrid deep learning architecture combines CNNs, BiLSTMs, and multi-head self-attention techniques. CNN layers captured spatial features from overlapping ECG data frames to identify localized waveform anomalies. Since BiLSTM layers capture bidirectional temporal interactions, the model can understand dynamic cardiac rhythm fluctuations. Multi-Head Attention increases classification and clinical interpretation by adaptively prioritizing important temporal segments. Strongness, convergence, and generalization are improved by residual connections, layer normalization, dropout regularization, and sliding-window segmentation. On the MIT-BIH Malignant VF and Normal Sinus Rhythm datasets, the model had 98.03% accuracy, 0.97 precision and recall scores, and near-perfect F1-scores with just two misclassifications out of 152 test examples. Better than usual CNN models. These findings suggest the program can predict VF in real time with clinical significance. With high detection performance, interpretability, responsiveness, and energy economy, the suggested system advances proactive, customized arrhythmia treatment employing wearable or implanted monitoring technologies.</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

Design and Development of Early Prediction Algorithm for Ventricular Fibrillation Using Novel Hybrid Deep Learning Model

  • 期刊论文

  • 发表日期：

    2026年2月2日

查看全文
  • Jaya Lakshmi Athukuri,
  • D. Haripriya

摘要

Ventricular Fibrillation (VF), a deadly cardiac arrhythmia, kills many unexpectedly, making early prediction crucial. The discovery plugs a gap in bespoke prediction systems that lack generalizability across patient profiles and real-time interpretability. To overcome these limitations, a novel hybrid deep learning architecture combines CNNs, BiLSTMs, and multi-head self-attention techniques. CNN layers captured spatial features from overlapping ECG data frames to identify localized waveform anomalies. Since BiLSTM layers capture bidirectional temporal interactions, the model can understand dynamic cardiac rhythm fluctuations. Multi-Head Attention increases classification and clinical interpretation by adaptively prioritizing important temporal segments. Strongness, convergence, and generalization are improved by residual connections, layer normalization, dropout regularization, and sliding-window segmentation. On the MIT-BIH Malignant VF and Normal Sinus Rhythm datasets, the model had 98.03% accuracy, 0.97 precision and recall scores, and near-perfect F1-scores with just two misclassifications out of 152 test examples. Better than usual CNN models. These findings suggest the program can predict VF in real time with clinical significance. With high detection performance, interpretability, responsiveness, and energy economy, the suggested system advances proactive, customized arrhythmia treatment employing wearable or implanted monitoring technologies.