Synchroextracting chirplet transform-based epileptic seizures detection using EEG. (July 2021)

• Record Type: Journal Article
• Authors: Jiang, Yun; Chen, Wanzhong; Li, Mingyang; Zhang, Tao; You, Yang
• Is Part Of: Biomedical signal processing and control. Volume 68(2021)
• Publication Date: 2021-07
• Subjects: EEG -- Time-frequency analysis -- Synchroextracting chirplet transform -- IF estimation -- Seizure detection; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2021.102699
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: Synchroextracting chirplet transform is exploited for epileptic EEG signal representation. The accurate TF trajectory obtained by SECO can fully explore the effective information of the epilepsy signal. Only the TF coefficients most relevant to the IF trajectory are extracted to remove the irrelevant noise interference. Satisfied performance indicators are both obtained using different EEG datasets including Bonn and CHB-MIT. The strong generalization ability and portability of the proposed method guarantee the application in clinical. Abstract: With the development of computer technology, computer-aided medical diagnosis based on signal processing is trending up. In recent years, the detection of epilepsy electroencephalogram (EEG) signals for clinical application has attracted extensive attention from many researchers. In this paper, a novel epilepsy classification model is proposed based on the time-frequency (TF) analysis method named synchroextracting chirplet transform (SECT). The energy concentrated TF representation of the signal is first obtained benefit by the introduced parameter chirp rate. Then the instantaneous frequency (IF) trajectory is estimated by extracting the TF coefficients most relevant to the signal through the synchroextracting chirplet operator (SECO). The original signal can be simple and effectively reconstructed using the accurate IF trajectory with anti-noise interference ability. Singular value decomposition is performed on the TF matrix to obtain the feature vectors with strong characterization capacity for different epilepsy behaviors, and the multiple classification tasks are finally completed with support vector machine (SVM) classifier. Two authoritative epilepsy datasets are employed in our work to verify the effectiveness of the proposed scheme. No less than 99.17 % accuracy, 99.33 % specificity, and 99.28 % sensitivity are achieved on a total of ten classification tasks that are closely related to the actual clinical applications in the Bonn dataset; an average of 99.29 % accuracy and 0.97 matthews correlation coefficient (MCC) are obtained among all subjects in the long-term EEG database CHB-MIT. The noise robustness and portability of our proposal are verified through the stable experimental results. The superior detection accuracy also provides clinicians with auxiliary diagnosis.