Abstracts

Rationale: The value of Ictal-SPECT in the pre-surgical evaluation of epilepsy patients is well established. Among ictal-SPECT processing methods SISCOM (Subtraction Ictal SPECT Coregistered to MRI) is the most widely accepted and extensively validated. However, other approaches such as ISAS (Ictal-Interictal Subtraction Analysis by Statistical Parametric Mapping) offer the objectivity of statistical measures. ISAS derives the statistical significance of a patient s SPECT activity through a voxelwise comparison to pooled and parameterized data from normal subjects. We sought to compare the inter-rater reliability of these two methods to visual analysis of raw SPECT data in terms of localization and study diagnostic conclusiveness. Methods: Ictal and interictal SPECT scans were obtained from 67 consecutive patients as part of their epilepsy pre-surgical evaluation. These raw scan pairs were each further processed using SISCOM and ISAS methods. A panel of 3 blinded experienced reviewers from different institutions (ALP, KHL, BCO) evaluated each patient s SPECT data in 3 forms; raw unprocessed ictal-interictal pair, with SISCOM processing, and with ISAS processing. Each reviewer was presented scans in a 16 axial slice (4x4) configuration and asked to identify 1) the location of the most significant SPECT activity within one of 30 pre-specified brain regions, and 2) the overall diagnostic conclusiveness of the study judged as definitely localizing , probably localizing , or not localizing. Proportions of inter-rater agreement were calculated, and inter-rater reliability among the 3 reviewers was determined for each of the 3 processing methods using Fleiss Kappa method. Results: In our study population, containing 20% temporal lobe and 80% extratemporal localization, the proportion of studies in which all 3 reviewers agreed on the same ROI as containing the most significant activity was RAW = 6% (95% CI, 0.3%-12%), SISCOM = 13% (5-22), and ISAS = 30% (19-40). All 3 reviewers agreed that no single ROI predominated as most significant in 37% (26-49), 22% (12-32), and 7% (1-14) respectively. Studies with total disagreement among all 3 reviewers on the most significant ROI were 9% (2-16), 18% (9-27), and 12% (4-20), respectively. The inter-rater reliability (?) was 0.36, 0.42, and 0.51 respectively when not localized was included as a 31st ROI. The proportion of studies judged by all 3 reviewers as localized or probably localized was 10% (3-18), 25% (15-36), and 40% (28-52), respectively. Conclusions: Ictal-interictal SPECT, when processed using the ISAS method, yields better localization agreement, fewer non-localized scans, and less reviewer total disagreement, than those with SISCOM processing. SISCOM was shown to be superior to raw scans in most but not all of these same measures. This ISAS>SISCOM>RAW agreement trend was also reflected in the inter-rater reliability (?) of the three methods. Finally, ISAS tended to more often produce scans judged as diagnostically conclusive, i.e. localizing, than SISCOM, and much more often than raw SPECT.