Scholarlink

<p>Epilepsy is a common neurological condition, characterized by sudden and unpredictable disturbances in brain activity. Conventional antiepileptic medications fail to control seizures in nearly one-third of patients, highlighting the need for alternative approaches to lessen the disease’s impact on patients and caregivers. While electroencephalography is the primary diagnostic tool, epilepsy is also associated with alterations in the autonomic nervous system. Recent studies suggest that electrocardiogram signals may reveal preictal changes in ANS activity. In this study, seizure prediction was performed offline by analyzing heart rate variability derived from video-EEG recordings of 13 hospitalized patients (including 34 seizures) at the Neurology and Neurophysiology Unit, University of Siena, Italy. After preprocessing, R-wave peaks were identified in the ECG using the Pan-Tompkins algorithm, and RR intervals were calculated. The resulting RR interval series was used to generate a two-dimensional Poincaré representation. From this, five time series were constructed, and features were extracted from temporal, spectral, time frequency, and nonlinear domains. Three preictal windows (5, 15, and 30 min) were evaluated. The features were classified using machine learning models, including support vector machines, k-nearest neighbors, and random forest. Results showed that the 5-min preictal window achieved the highest accuracy, with nearly 100% sensitivity and the lowest false alarm rate. The 15-min interval showed slightly lower performance but maintained strong accuracy and AUC, making it suitable for early-warning systems. These findings indicate that multidomain feature extraction combined with nonlinear classifiers and HRV dynamics via the Poincaré approach provides an effective complementary method for seizure prediction in non–real-time monitoring systems.</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

A Novel Non-Real-Time Algorithm for Epileptic Seizure Prediction Using Features Extracted from Multiple Time Series Derived from the Two-Dimensional Poincaré Plot of RR Intervals

  • 期刊论文

  • 发表日期：

    2026年3月3日

查看全文
  • Pardis Goodarzi,
  • Keivan Maghooli,
  • Nader Jafarnia Dabanloo,
  • Fardad Farokhi

摘要

Epilepsy is a common neurological condition, characterized by sudden and unpredictable disturbances in brain activity. Conventional antiepileptic medications fail to control seizures in nearly one-third of patients, highlighting the need for alternative approaches to lessen the disease’s impact on patients and caregivers. While electroencephalography is the primary diagnostic tool, epilepsy is also associated with alterations in the autonomic nervous system. Recent studies suggest that electrocardiogram signals may reveal preictal changes in ANS activity. In this study, seizure prediction was performed offline by analyzing heart rate variability derived from video-EEG recordings of 13 hospitalized patients (including 34 seizures) at the Neurology and Neurophysiology Unit, University of Siena, Italy. After preprocessing, R-wave peaks were identified in the ECG using the Pan-Tompkins algorithm, and RR intervals were calculated. The resulting RR interval series was used to generate a two-dimensional Poincaré representation. From this, five time series were constructed, and features were extracted from temporal, spectral, time frequency, and nonlinear domains. Three preictal windows (5, 15, and 30 min) were evaluated. The features were classified using machine learning models, including support vector machines, k-nearest neighbors, and random forest. Results showed that the 5-min preictal window achieved the highest accuracy, with nearly 100% sensitivity and the lowest false alarm rate. The 15-min interval showed slightly lower performance but maintained strong accuracy and AUC, making it suitable for early-warning systems. These findings indicate that multidomain feature extraction combined with nonlinear classifiers and HRV dynamics via the Poincaré approach provides an effective complementary method for seizure prediction in non–real-time monitoring systems.