Shaheed Bhagat Singh State University, Ferozepur, Punjab, India.
Malout Institute of Management & Information Technology, Malout, Punjab, India.
Phys Eng Sci Med. 2022 Mar;45(1):261-272. doi: 10.1007/s13246-022-01111-9. Epub 2022 Feb 15.
Epilepsy is a chronic neurological disorder that involves abnormal electrical signal patterns of the brain called seizures. The brain's electrical signals can be recorded using an electroencephalogram (EEG). EEG recordings can be used to monitor complex and non-stationary signals produced by the brain for detecting epilepsy seizures. Machine learning (ML) methods have been successfully applied in different domains to accurately classify the patterns based upon dataset features. However, ML methods are unable to analyze the raw EEG signals. Appropriate features must be extracted from EEG recordings for detecting epilepsy seizures using signal processing methods. This work proposes an intelligent system by integrating variational mode decomposition (VMD) and Hilbert transform (HT) method for extracting useful features from EEG signals and stacked neural network (NN) method for detecting epilepsy seizures. VMD method decomposers EEG signals into intrinsic mode functions, followed by the HT method for extracting features from EEG signals. The stacked-NN approach is applied for detecting epilepsy seizures using extracted features. The performance of the proposed system is validated using benchmark datasets for epilepsy seizure detection provided by Bonn University and, Neurology and Sleep Centre, New Delhi (NSC-ND). The performance of the proposed system is compared with other ML methods and state of the art approaches in the field. The reported results demonstrate that the proposed system can detect up to 100% accurate epilepsy seizures using NSC-ND data set and up to 99% accurate epilepsy seizures using Bonn university dataset. The comparative results also demonstrate the better performance of the proposed system over other ML methods and existing approaches for detecting epilepsy seizures. The remarkable performance of the proposed system can help neurological experts to detect epilepsy seizures accurately using EEG signals and can be embedded into the real-time diagnosis of the disease.
癫痫是一种慢性神经系统疾病，涉及大脑异常电信号模式，称为癫痫发作。大脑的电信号可以通过脑电图（EEG）记录。脑电图记录可用于监测大脑产生的复杂和非平稳信号，以检测癫痫发作。机器学习（ML）方法已成功应用于不同领域，以根据数据集特征准确分类模式。然而，ML 方法无法分析原始 EEG 信号。必须使用信号处理方法从 EEG 记录中提取适当的特征，以检测癫痫发作。这项工作提出了一种通过集成变分模态分解（VMD）和希尔伯特变换（HT）方法从 EEG 信号中提取有用特征的智能系统，以及用于检测癫痫发作的堆叠神经网络（NN）方法。VMD 方法将 EEG 信号分解为固有模态函数，然后使用 HT 方法从 EEG 信号中提取特征。堆叠-NN 方法用于使用提取的特征检测癫痫发作。该系统的性能使用波恩大学和新德里神经病学和睡眠中心（NSC-ND）提供的癫痫发作检测基准数据集进行验证。将该系统的性能与其他 ML 方法和该领域的最新方法进行比较。报告的结果表明，该系统可以使用 NSC-ND 数据集检测高达 100%的准确癫痫发作，使用波恩大学数据集检测高达 99%的准确癫痫发作。比较结果还表明，该系统在检测癫痫发作方面的性能优于其他 ML 方法和现有的方法。该系统的卓越性能可以帮助神经科专家使用 EEG 信号准确检测癫痫发作，并可以嵌入到疾病的实时诊断中。
