Purpose: To examine whether or not pattern analysis of magnetic resonance volumetric (MRVol) and proton magnetic resonance spectroscopic imaging (1H-MRSI) data would enable (a) the accurate lateralization of temporal lobe epilepsy (TLE) and (b) the discrimination of TLE from extratemporal epilepsy (E-TLE).
Methods: For lateralization analysis, we used data from 150 nonforeign tissue lesional TLE patients [88 left-sided (L-TLE), 46 right-sided (R-TLE), and 16 bilateral (Bi-TLE)]. For the discrimination of TLE from E-TLE, we used data from 174 patients (145 with unilateral TLE, 14 with unilateral E-TLE, and 15 with widespread epileptogenic zones involving both the TL and extra-TL regions-multilobar epilepsy). A series of "leave-one-out" cross-validated linear discriminant analyses were performed using the MRVol and 1H-MRSI data sets to lateralize TLE and discriminate it from E-TLE.
Results: Lateralization: The leave-one-out linear discriminant analyses were able to correctly lateralize (with a posterior probability >0.50) 120 (90%) of the 134 L-TLE and R-TLE patients. Imposing higher posterior probability (>0.95) increased accuracy of lateralization to 98%, with only two discordant cases who underwent surgery on the side of electroencephalogram, and both had bad outcome. Discrimination: the leave-one-out linear discriminant analyses were able to correctly classify (with a posterior probability >0.50) 142 (89%) of the 159 TLE and E-TLE patients. Accuracy increased slightly as higher posterior probability cutoffs were imposed, with fewer patients being classified.
Conclusions: Pattern analysis of 1H-MRSI and MRVol data can accurately lateralize TLE. Discriminating TLE from E-TLE was less accurate, probably due to the presence of temporal lobe damage in some patients with E-TLE reflecting dual pathology.