Abstracts

An MR-Based Bayes Classifier for the Prediction of Surgical Outcome in Temporal Lobe Epilepsy Patients.

Abstract number : F.04
Submission category :
Year : 2000
Submission ID : 3196
Source : www.aesnet.org
Presentation date : 12/2/2000 12:00:00 AM
Published date : Dec 1, 2000, 06:00 AM

Authors :
Samson B Antel, Li Min Li, Fernando Cendes, Andre Olivier, Frederick Andermann, Francois Dubeau, Robert E Kearney, Rajjan Shinghal, Douglas L Arnold, Montreal Neurological Institute, Montreal, PQ, Canada; Montreal Neurological Hosp and Institute, Montreal

RATIONALE: Conventional pre-surgical investigation, based on video-EEG, is costly and inconvenient, requiring weeks of hospitalization. Magnetic resonance imaging (MRI) and magnetic resonance spectroscopic imaging (MRSI) allow rapid, in-vivo measurement of structural and metabolic information within the brain, and provide markers correlating with lateralization and outcome. Our aim was to use MR data to develop a quantitative model to predict surgical outcome. METHODS: We examined 81 patients with non-foreign tissue lesional TLE. Prior to surgery, we measured ipsilateral and contralateral values for four attributes: amygdaloid and hippocampal volume in T1 MRI, and the ratio of N-acetyl-aspartate (NAA) to creatine (Cr) in the mid and posterior temporal regions using MRSI (TE/TR=272/2000, 1.5T). For each attribute, we calculated an asymmetry score defined as (ipsilateral value - contralateral value). A leave-one-out naive Bayes classifier was developed to predict surgical outcome, based on inputs chosen from the above attributes. Outcome was defined as worthwhile improvement (Engel's Classes I-III, n=65) or non-worthwhile improvement (Engel's Class IV, n=16). RESULTS: The naive Bayes classifier correctly predicted outcome for 95% (62/65) of worthwhile improvement patients, and 75% (12/16) of non-worthwhile improvement patients. Predictive value was 94% (62/66) for worthwhile improvement and 80% (12/15) for non-worthwhile improvement. Highest information gain was provided by NAA/Cr level in the contralateral posterior temporal region and degree of hippocampal volume asymmetry. Asymmetry scores for all attributes were significantly more ipsilaterally weighted in the worthwhile improvement group. Contralateral hippocampal volume and NAA/Cr level in the contralateral posterior temporal region were significantly lower in the non-worthwhile improvement group. CONCLUSIONS: Combined MRI and MRSI can be used to predict surgical outcome. Prolonged video-EEG may therefore be unnecessary for many surgical candidates. Indicators of lateralization and degree of contralateral involvement are important markers.