Search

• Phase amplitude coupling shows distinct patterns of cross frequency interactions in healthy controls and temporal lobe epilepsy patients even during resting state records. • SVM shows high classification potential to classify healthy controls from temporal lobe epilepsy patients and to lateralize epileptic focus in temporal lobe epilepsy using phase amplitude coupling. • Low gamma (30–80 Hz) interactions with lower frequency band (1–8 Hz) can identify hemisphere with epileptic focus in unilateral drug resistant temporal lobe epilepsy in resting state MEG even in the lack of ictal or interictal discharges. • Phase amplitude coupling in resting state records can supplement video-EEG in the pre-surgical evaluation of drug resistant unilateral temporal lobe epilepsy. Lateralization of seizure focus in temporal lobe epilepsy (TLE) is a prime step in pre-surgical evaluation requiring prolonged seizure monitoring using video EEG and manual inspection of recordings. This study uses phase amplitude coupling (PAC) in resting state magnetoencephalography to automatically lateralize TLE focus. Fifty-four patients with drug resistant TLE and 21 healthy controls who underwent MEG were considered for the study. Classification was carried out for PAC calculated for source transformed resting state of controls vs left TLE (LTLE)/right TLE (RTLE) and LTLE vs RTLE between beta, low-gamma and high-gamma as high frequency (HF) bands and low frequency (LF) 1–13 Hz, with decision tree (DT), support vector machines (SVM) and naïve Bayes with feature selection by chi-square test. Further, lateralization classification was also calculated with LF sub-bands (delta, theta, alpha). PAC was higher in the TLE compared to controls. LTLE and RTLE showed differences in low gamma-alpha and high gamma-delta coupling (p < 0.05). Accuracy was highest with SVM between controls and LTLE in the low gamma-LF (92.92%, AUC-1), between controls and RTLE in DT and SVM (93.54%, AUC-0.97, 1) in the low gamma-LF band and in low gamma-delta band in SVM (92.04%, AUC-1) between LTLE and RTLE. PAC shows distinct patterns of coupling in each subject group. Feature selection showed involvement of major network hubs and resting state networks. SVM showed best classification potential in the low gamma band. PAC in resting state MEG can supplement pre-surgical evaluation in drug resistant TLE. [ABSTRACT FROM AUTHOR]