Neural network analysis of preoperative variables and outcome in epilepsy surgery

Citation
Je. Arle et al., Neural network analysis of preoperative variables and outcome in epilepsy surgery, J NEUROSURG, 90(6), 1999, pp. 998-1004
Citations number
29
Categorie Soggetti
Neurology,"Neurosciences & Behavoir
Journal title
JOURNAL OF NEUROSURGERY
ISSN journal
00223085 → ACNP
Volume
90
Issue
6
Year of publication
1999
Pages
998 - 1004
Database
ISI
SICI code
0022-3085(199906)90:6<998:NNAOPV>2.0.ZU;2-C
Abstract
Object. Because appropriate patient selection is essential for achieving su ccessful outcomes after epilepsy surgery, the need for more robust methods of predicting postoperative seizure control has been coated. Standard multi variate techniques have been only 75 to 80% accurate in this regard. Recent use of artificial intelligence techniques, including neural networks, for analyzing multivariate clinical data has been successful in predicting medi cal outcome. Methods. The authors applied neural network techniques to 80 consecutive pa tients undergoing epilepsy surgery in whom data on demographic, seizure, op erative, and clinical variables to predict postoperative seizures were coll ected. Neural networks could be used to predict postoperative seizures in up to 98 % of cases. Student's t-tests or chi-square analysis performed on individua l variables revealed that only the preoperative medication index was signif icantly different (p = 0.02) between the two outcome groups. Six different combinations of input variables were used to train the networks . Neural ne twork accuracies differed in their ability to predict seizures: using all d ata (96%): all data minus electroencephalography concordance and operative side (93%); all data except intra- or postoperative variables such as tissu e pathological category (98%); all data excluding pathological category, in telligence quotient (IQ) data, and Wada results (84%); only demographics an d tissue pathological category (65%); and only IQ data (63%). Conclusions. Analysis of the results reveals that several networks that an trained with the usual accepted variables characterizing the typical evalua tion of epilepsy patients can predict postoperative seizures with greater t han 95% accuracy.