Abstracts

Rationale: Factor analysis (FA) is a common machine learning method used to find latent interdependencies. Even though principal component analysis (PCA) and FA are data reduction techniques, PCA is a linear combination of uncorrelated variables with high variance while FA reveals latent variables that cause the observable variables to covary. We applied factorial analysis to intracranial EEG data in the early ictal phase in an attempt to delineate the seizure onset zone. Methods: A 12 channel, 10,000 data point EEG was simulated in MATLAB version 9.5.0 (R2018b) and FA was applied to the data. The first 2 channels were modelled to be highly correlated. The first 2 loadings for each of the 12 channels were plotted after FA in a scatter plot. The procedure was repeated 25 times and scatter plots were reviewed for proximity of the first 2 channels in space. Two patients with intractable focal epilepsy who underwent detailed pre-surgical evaluation at the University of Pittsburgh Medical center were selected and the demographic, imaging, and electrophysiological data were reviewed. Patient 1 underwent resective surgery after SEEG evaluation and remained seizure-free for one year (Engel Class1). Patient 2 underwent invasive EEG evaluation and was presented in the weekly multidisciplinary epilepsy conference. FA was applied to IEEG data of both patients in the early ictal phase of typical seizures (within 5 seconds of seizure onset) and in the pre-ictal phase (less than 30 minutes prior to the seizure onset). The first and second loadings of each channel was plotted and reviewed. Results: FA in simulated EEG data showed that highly correlated channels (1 and 2) remained in close proximity in the scatter plot with loadings as axis, compared to channels 3 to 12 in all the 25 trials. In FA of the two patients, the channels forming a separate cluster in the scatter plot of loadings in the early ictal phase correlated with the seizure onset zone in patient 1 and were deemed part of the channels of interest for possible resection in patient 2. Similar clusters were noted during FA of several pre-ictal epochs (5 seconds duration) in both patients. Conclusions: Factorial analysis of IEEG in the early ictal and pre-ictal phases can indicate the seizure onset zone. Detailed studies in larger cohorts can assess the efficacy of this methodology in the surgical evaluation of focal epilepsy. Funding: No funding