Epileptic seizure detection in EEG signals using tunable-Q factor wavelet transform and bootstrap aggregating
Epileptic seizure detection in EEG signals using tunable-Q factor wavelet transform and bootstrap aggregating
复制标题
DOI：
10.1016/j.cmpb.2016.09.008
复制
发表时间：
2016-12-01
影响因子：
6.1
通讯作者：
Zhang, Yanchun
中科院分区：
文献类型：
--
作者：
Hassan, Ahnaf Rashik;Siuly, Siuly;Zhang, Yanchun
背景和目的：癫痫发作的检测传统上是由专家临床医生基于对脑电信号的视觉观察来完成的。这一过程耗时、繁重、依赖昂贵的人力资源，而且容易出现错误和偏见。另一方面，在癫痫研究中，人工检测不适合处理大数据集。一种计算机化的癫痫发作识别方案可以根治上述问题，帮助临床医生，并有利于癫痫研究。方法：提出了一种新的基于脑电信号的Tunable-Q因子小波变换(Tqwt)和Bootstrap聚集(Bging)的癫痫自动诊断方案。到目前为止，据作者所知，这是第一次将Tqwt域的光谱特征与袋装相结合用于癫痫发作的识别。首先利用Tqwt将脑电信号段分解为多个子带。然后从Tqwt子带中提取各种光谱特征。通过统计测量和图形分析确定了光谱特征在Tqwt域中的适宜性。之后，采用袋法对癫痫发作进行分类。在本研究中，还研究了袋化在所提出的检测方案中的有效性。研究了不同的TQWT和袋装参数对包装效果的影响。并确定了这些参数的最优选择。使用公开可用的基准脑电图库对发作间期(无发作间期)、发作(发作)与健康、发作与非发作、发作与发作、发作与健康等不同分类案例的性能进行了研究。结果：与最新的算法相比，本文提出的癫痫检测方法在敏感性、特异性和准确性方面都具有较好的性能。结论：本文提出的癫痫检测方法可以减轻医学专业人员通过视觉检查分析大量数据的负担，加快癫痫的诊断速度，有利于癫痫的研究。(C)2016爱思唯尔爱尔兰有限公司。保留所有权利。
Background and objective: Epileptic seizure detection is traditionally performed by expert clinicians based on visual observation of EEG signals. This process is time-consuming, burdensome, reliant on expensive human resources, and subject to error and bias. In epilepsy research, on the other hand, manual detection is unsuitable for handling large datasets. A computerized seizure identification scheme can eradicate the aforementioned problems, aid clinicians, and benefit epilepsy research.Methods: In this work, a new automated epilepsy diagnosis scheme based on Tunable-Q factor wavelet transform (TQWT) and bootstrap aggregating (Bagging) using Electroencephalogram (EEG) signals is proposed. Until now, this is the first time spectral features in the TQWT domain in conjunction with Bagging are employed for epilepsy seizure identification to the best of the authors' knowledge. At first, we decompose the EEG signal segments into sub-bands using TQWT. We then extract various spectral features from the TQWT sub-bands. The suitability of spectral features in the TQWT domain is established through statistical measures and graphical analyses. Afterwards, Bagging is employed for epileptic seizure classification. The efficacy of Bagging in the proposed detection scheme is also studied in this research. The effects of various TQWT and Bagging parameters are investigated. The optimal choices of these parameters are also determined. The performance of the proposed scheme is studied using a publicly available benchmark EEG database for various classification cases that include inter-ictal (seizure-free interval), ictal (seizure) and healthy; seizure and non-seizure; ictal and inter-ictal; and seizure and healthy.Results: In comparison with the state-of-the-art algorithms, the performance of the proposed method is superior in terms of sensitivity, specificity, and accuracy.Conclusion: The seizure detection method proposed herein therefore can alleviate the burden of medical professionals of analyzing a large bulk of data by visual inspection, speed-up epilepsy diagnosis and benefit epilepsy research. (C) 2016 Elsevier Ireland Ltd. All rights reserved.