Scholarlink

<p>Epileptic seizures can happen for a variety of reasons, including skull fractures, genetic predisposition, tumours, and other things that may make them more likely to happen. Consistency in accurate seizure detection continues to be a clinical challenge. We have developed a means by which we can optimize patient-specific electrodes dynamically in real time through using XGBoost (XGB) band-power analysis to select the best channels uniquely for each patient; this is a dramatic difference from the fixed montages traditionally used in the past. our ConvLSTM architecture’s channel importance mapping provides unprecedented clinical interpretability, revealing seizure onset zones consistent with known propagation networks.This integrated approach validated across diverse Children Hospital Boston (CHB-MIT) patients achieves macro F1-score of 93.4% and a low false alarm rate of 0.13/hr while maintaining real time processing capabilities. Notably, patients like chb16 anomaly F1 96.7% exemplify the model’s potential for reliable long-term monitoring.The study successfully demonstrates an interpretable anomaly detection framework for epileptic seizure recognition in long, imbalanced EEG recordings a critical clinical challenge often overlooked in prior work.</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 anomaly detection using XGBoost and ConvLSTM model

  • 期刊论文

  • 发表日期：

    2026年5月12日

查看全文
  • Wenjuan Wu,
  • Jiafei Dai,
  • Mohsin Hasan,
  • Xudong Gu

摘要

Epileptic seizures can happen for a variety of reasons, including skull fractures, genetic predisposition, tumours, and other things that may make them more likely to happen. Consistency in accurate seizure detection continues to be a clinical challenge. We have developed a means by which we can optimize patient-specific electrodes dynamically in real time through using XGBoost (XGB) band-power analysis to select the best channels uniquely for each patient; this is a dramatic difference from the fixed montages traditionally used in the past. our ConvLSTM architecture’s channel importance mapping provides unprecedented clinical interpretability, revealing seizure onset zones consistent with known propagation networks.This integrated approach validated across diverse Children Hospital Boston (CHB-MIT) patients achieves macro F1-score of 93.4% and a low false alarm rate of 0.13/hr while maintaining real time processing capabilities. Notably, patients like chb16 anomaly F1 96.7% exemplify the model’s potential for reliable long-term monitoring.The study successfully demonstrates an interpretable anomaly detection framework for epileptic seizure recognition in long, imbalanced EEG recordings a critical clinical challenge often overlooked in prior work.