Scholarlink

<p>The analysis of electrocardiogram (ECG) signals is essential for detecting arrhythmias such as bradycardia, ventricular tachycardia, and atrial fibrillation. This study utilizes MIMIC-IV-ECG dataset, containing over 800,000 recordings, to assess the effectiveness of spiking neural networks (SNNs) in arrhythmia finding. Various deep learning architectures including hybrid, leaky integrate-and-fire networks, spiking convolutional neural networks (SCNNs), convolutional spiking neural networks, S-RNN, and LSTM-SNN, are trained using key ECG features, with performance enhanced through data augmentation and feature engineering. Our findings identified the innovative SCNN demonstrate outstanding arrhythmia classification ability at a highly to notable 97% accuracy; while hybrid models like norse-hybrid and dense-spike show their own capabilities by incorporating traditional deep learning architectures with the concept of spiking neurons and offer additional performance improvements. These findings highlight the potential of neuromorphic computing for ECG analysis, with future work focusing on real-time processing and clinical scalability.</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

Enhanced arrhythmia detection using spiking neural networks: an in-depth analysis of ECG data from the MIMIC-IV clinical database

  • 期刊论文

  • 发表日期：

    2025年9月5日

查看全文
  • Gunjan Verma,
  • Honey Gocher,
  • Sweety Verma

摘要

The analysis of electrocardiogram (ECG) signals is essential for detecting arrhythmias such as bradycardia, ventricular tachycardia, and atrial fibrillation. This study utilizes MIMIC-IV-ECG dataset, containing over 800,000 recordings, to assess the effectiveness of spiking neural networks (SNNs) in arrhythmia finding. Various deep learning architectures including hybrid, leaky integrate-and-fire networks, spiking convolutional neural networks (SCNNs), convolutional spiking neural networks, S-RNN, and LSTM-SNN, are trained using key ECG features, with performance enhanced through data augmentation and feature engineering. Our findings identified the innovative SCNN demonstrate outstanding arrhythmia classification ability at a highly to notable 97% accuracy; while hybrid models like norse-hybrid and dense-spike show their own capabilities by incorporating traditional deep learning architectures with the concept of spiking neurons and offer additional performance improvements. These findings highlight the potential of neuromorphic computing for ECG analysis, with future work focusing on real-time processing and clinical scalability.