A probabilistic approach for lateralization of seizure onset zone in drug-resistant epilepsy with bilateral cerebral pathology.

ABSTRACT

BACKGROUND: Lateralization of seizure-onset zone (SOZ) during electroencephalography (EEG) monitoring in people with bilateral potentially epileptogenic lesions is important to facilitate clinical decision making for resective surgery. METHODS: We develop two Bayesian approaches for estimating the number of consecutive ipsilateral seizures required to lateralize the SOZ to a given lower limit of 95% credible interval (LLI, assuming continuous prior distribution), or to a given posterior probability (assuming mixture of discrete and continuous prior probabilities). RESULTS: With estimation approach, if both the cerebral hemispheres are a priori equi-probable to contain SOZ, then using Jeffrey's prior, a minimum of 9, 18, and 38 consecutive ipsilateral seizures will yield an LLI of 0.81, 0.90, and 0.95 respectively. If one of the hemisphere is a priori more likely to have SOZ, then prior beta distributions with α=3, ß=2, and α=4, ß=3 will require a minimum of 18 and 24 consecutive ipsilateral seizures to yield an LLI of 0.80. Contrariwise, the testing approach allows approximation of the number of consecutive ipsilateral seizures to lateralize the SOZ depending on an estimate of prior probability of lateralized SOZ, to a desired posterior probability. For a prior probability of 0.5, using uniform prior, mixture model will require 7, 17, and 37 consecutive ipsilateral seizures to lateralize the SOZ with a posterior probability of 0.8, 0.9, and 0.95 respectively. CONCLUSION: While the reasoning presented here is based on probability theory, it is hoped that it may help clinical decision making and stimulate further validation with actual clinical data.