Modularity-Based Detection of Ripples in Scalp EEG

Ripples (80-250Hz) are promising markers of epileptogenic activity, but the diagnostic value of ripples in scalp EEG remains debated. In this study, we propose an unsupervised, cluster-based method to detect candidate ripples in scalp EEG and sort ripples according to their morphology and information content in the time-frequency domain. We also correlate the presence of ripples to the presence of interictal spikes, which are clinically recognized markers of epileptogenic activity. Our method combines feature-based agglomerative clustering and correlation-based community detection and was tested on scalp EEG from 3 children with epilepsy (age: 10±1 [mean ± SD], 2 male, 1 female). For each patient, one epoch of EEG during wakefulness and one epoch during sleep (stage N2-N3) were considered (wakefulness: 12.57±3.39 min; sleep: 14.68±0.49 min, mean ± SD). The proposed method showed high specificity in detecting ripples while rejecting artifacts and resulted in a minimal set of ripple templates that are consistent across patients and sleep condition. Also, the rate of ripples was higher in EEG channels that presented spikes (0.38±0.07 versus 0.24±0.07 ripples/min [mean ± SD]). Altogether, results indicate that morphology and spectral content of scalp ripples may be patient-independent and specific to the epileptogenic activity, which suggest scalp ripples as viable markers for noninvasive epilepsy diagnosis.