A multistructural imaging marker for non-invasive lateralization of temporal lobe epilepsy

This study investigates the predictive power of feature sets extracted from different brain structures for lateralization of the epileptogenic focus in mesial temporal lobe epilepsy (mTLE) patients based on imaging features. To this end, volumes of multiple brain structures are extracted from preoperative images of 68 unilateral mTLE patients. Our data set consists of 54 patients with visually observable (positive) hippocampus sclerosis (HS-P) and 14 patients without visible (negative) hippocampus sclerosis (HS-N). Exploiting different structural volumes, contributions of the structures and their correlations to each other is evaluated by data mining techniques. After removing redundant correlated structures, a minimum set of structures is obtained as a mTLE lateralization marker. Our experiments, using volumes of hippocampus, amygdala, and thalamus show a correct lateralization rate of 98.5%. This ternary-structural marker also shows 100% and 93% mTLE lateralization accuracy respectively for HS-P and HS-N groups. In conclusion, the proposed marker improves decision-making of surgical resection especially for HS-N group and may reduce the need for implantation of intracranial monitoring electrodes.