Scholarlink

<p>The prevalence of research on harmful brain activity has increased, especially since the standardization of electroencephalography (EEG) terminologies. A continual lack of specialists leads to considerable distress and increased mortality rates among critically ill patients. An automated system that can detect and classify harmful brain activities is crucial, thereby improving patient safety and potentially saving lives. This research presents a new pipeline for classifying harmful brain activities. Multiple features were used for feature fusion, which was implemented via a dual 1D convolutional neural network (CNN) model. A series of experiments was conducted to demonstrate the robustness of the proposed pipeline by using the Harvard Medical School (HMS) dataset. An ablation analysis and explainable AI were utilized to illustrate the robustness of the proposed pipeline. A feature-level fusion scheme for classifying seizures and seizure-like patterns is presented in this study. The best model proposed in this study, i.e., model 8, achieved an accuracy of 98.14% with a loss of 0.05, whereas the 10-fold cross-validation results obtained for the same model yielded an accuracy of 99%.</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

EEG-based harmful brain activity classification using deep learning and feature fusion

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月6日

查看全文
  • Zaib Unnisa,
  • Arfan Jaffar,
  • Sheeraz Akram,
  • IrfanUd Din,
  • Khalil Khan,
  • Sohail Masood Bhatti

摘要

The prevalence of research on harmful brain activity has increased, especially since the standardization of electroencephalography (EEG) terminologies. A continual lack of specialists leads to considerable distress and increased mortality rates among critically ill patients. An automated system that can detect and classify harmful brain activities is crucial, thereby improving patient safety and potentially saving lives. This research presents a new pipeline for classifying harmful brain activities. Multiple features were used for feature fusion, which was implemented via a dual 1D convolutional neural network (CNN) model. A series of experiments was conducted to demonstrate the robustness of the proposed pipeline by using the Harvard Medical School (HMS) dataset. An ablation analysis and explainable AI were utilized to illustrate the robustness of the proposed pipeline. A feature-level fusion scheme for classifying seizures and seizure-like patterns is presented in this study. The best model proposed in this study, i.e., model 8, achieved an accuracy of 98.14% with a loss of 0.05, whereas the 10-fold cross-validation results obtained for the same model yielded an accuracy of 99%.