Automated focal EEG signal detection based on third order cumulant function. (April 2020)

• Record Type: Journal Article
• Authors: Sharma, Rahul; Sircar, Pradip; Pachori, Ram Bilas
• Is Part Of: Biomedical signal processing and control. Volume 58(2020)
• Publication Date: 2020-04
• Subjects: EEG -- Non-focal and focal -- Higher-order statistics -- LSDA -- Kernal -- SVM; 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.2020.101856
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: The epileptic seizures are only cured by the epilepsy surgery. The epileptogenic zone can be characterized by focal EEG signals. A nonlinear HOS method is used to classify focal and non-focal EEG signals. The proposed algorithm acheived 99% classification accuracy on publically available database. Abstract: Epilepsy is a chronic neurological disorder which occurs due to recurrent seizures. The epilepsy surgery is the only cure of epileptic seizures as it cannot be controlled with medication. Hence, it becomes the primary task to localize the epileptogenic zone for successful epilepsy surgery. The epileptic surgical area can be recognized by the focal intracranial electroencephalogram (EEG) signals. In this paper, a nonlinear third-order cumulant has been proposed for the classification of the non-focal and focal intracranial EEG signals efficiently. The attributes are measured from the logarithm of the diagonal slice of third-order cumulant. It provides relevant subtle information about the nonlinear dynamics of EEG signals. A data reduction technique, locality sensitive discriminant analysis (LSDA), has been introduced to map the measured features at higher dimensional space and ranked them according to the probability of discrimination. The achieved results reveal that the ranked LSDA features with the support vector machine (SVM) classifier have yielded maximum classification accuracy of 99% on the Bern Barcelona EEG database. Thus, the proposed algorithm helps a clinician to localize the epileptogenic zone for successful brain surgery.