Abstracts

Determining whether a paroxysmal EEG event in the temporal lobe is normal or abnormal is difficult. Normal temporal lobe paroxysmal events such as exagerations of alpha activity, wicket spikes, and small sharp spikes can have an epileptiform appearance. This study uses a new implementation of independent component analysis (ICA) called multitaper ICA (mICA) to analyze temporal lobe paroxysmal EEG discharges. The mICA technique performs ICA multiple times with multiple window lengths on an EEG dataset. It is able to detect paroxysmal EEG events and it represents each event with its own independent component and scalp distribution., A database of 101 de-identified 30-second segments of routine EEGs was prepared. This database included 61 segments containing temporal paroxysmal discharges from 50 patients without of history of epilepsy which were judged to be normal but difficult to interpret. Also placed in this database were 40 segments containing abnormal epileptiform discharges from 29 patients with known epilepsy. These abnormal epileptiform discharges were chosen for being subtle due to their relatively low amplitude. The mICA technique was used to generate independent components (ICs) containing the 101 paroxysmal discharges. Various characteristics of the ICs and scalp distributions of these discharges were studied to find characteristics which differentiated normal from abnormal temporal discharges. Characteristics studied include the scalp distribution, the number of ICs detected by mICA for each discharge, and the peak spectral frequency of the ICs., Characteristics associated with abnormal epileptiform paroxysmal discharges included: (1) mICA detection of three or more ICs for the discharge (2) discharges which produced ICs longer than one second in duration (3) ICs with scalp distributions which were not bitemporal (4) ICs with scalp distributions in which the temporal lobes had opposite polarities (5) ICs with scalp distributions which did not involve the occipital region (6) ICs with scalp distributions which had a low weights polarity (7) ICs with scalp distributions which were very focal due to involving only two or three electrodes in a standard 10-20 montage (8) and ICs which had a peak spectral frequency in the delta range due to an associated slow wave. Using these eight characteristics, the mICA technique for identifying abnormal epileptiform discharges labeled the discharges as abnormal with a sensitivity of 71% and a specificity of 82%., The study demonstrates subtle differences between normal and abnormal paroxysmal temporal EEG discharges. This technique could be implemented in routine EEG interpretation by allowing the EEG reader to highlight a short portion of one channel of EEG for running mICA analysis on. This study could be improved by including characteristics of the waveform shape (such as half-wave analysis and/or wavelet analysis) and more detailed frequency spectrum analysis.,