Automated temporal lobe epilepsy and psychogenic nonepileptic seizure patient discrimination from multichannel EEG recordings using DWT based analysis. (August 2022)

• Record Type: Journal Article
• Authors: Fıçıcı, Cansel; Telatar, Ziya; Eroğul, Osman
• Is Part Of: Biomedical signal processing and control. Volume 77(2022)
• Publication Date: 2022-08
• Subjects: EEG -- Epilepsy detection -- Machine learning -- Psychogenic nonepileptic seizure -- Temporal lobe epilepsy; 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.2022.103755
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: TLE patient, PNES patient and healthy subject discrimination was realized using EEG recordings is in order to eliminate the misdiagnosis and long inspection time of EEG recordings. This automated system is also eligible for classifying EEG epochs as interictal, ictal, healthy and PNES. Abstract: Psychogenic nonepileptic seizure (PNES) and epileptic seizure resemble each other, behaviorally. This similarity causes misdiagnosis of PNES and epilepsy patients, thus patients suffering from PNES may be treated with antiepileptic drugs which can have various side effects. Furthermore, seizure is diagnosed after time consuming examination of electroencephalography (EEG) recordings realized by the expert. In this study, automated temporal lobe epilepsy (TLE) patient, PNES patient and healthy subject discrimination method from EEG signals is proposed in order to eliminate the misdiagnosis and long inspection time of EEG recordings. Also, this study provides automated approach for TLE interictal and ictal epoch classification, and TLE, PNES and healthy epoch classification. For this purpose, subbands of EEG signals are determined from discrete wavelet transform (DWT), then classification is performed using ensemble classifiers fed with energy feature extracted from the subbands. Experiments are conducted by trying two approaches for TLE, PNES and healthy epoch classification and patient discrimination. Results show that in the TLE, PNES and healthy epoch classification the highest accuracy of 97.2%, sensitivity of 97.9% and specificity of 98.1% were achieved by applying adaptive boosting method, and the highest accuracy of 87.1%, sensitivity of 86.0% and specificity of 93.6% were attained using random under sampling (RUS) boosting method in the TLE patient, PNES patients and the healthy subject discrimination.