Abstracts

Rationale: In absence of an MRI lesion, epilepsy surgery outcomes are markedly reduced. Ictal SPECT has proven to be a useful tool in presurgical evaluation of epilepsy surgery candidates, and subtraction of ictal and interictal SPECT with coregistration to MRI (O Brien et. al. 1998) has become a standard part of seizure onset localization. The specificity of SISCOM, however, is likely reduced by the physiological variability of interictal perfusion. To overcome this issue, methods using statistical parametric mapping (SPM) with a population of control subjects have been developed (ISAS, McNally et al., 2005; STATISCOM, Kazemi et al., 2010). Manual review of these scans, however, is time-consuming and subjective, and reviewers diagnostic decision can be influenced by results of other diagnostic modalities. Methods: group for statistical analysis. Hyperperfusion and hypoperfusion maps were created for each patient and reviewed by 3 reviewers blinded to processing methods and clinical details in custom software. Their consensus was used to determine the lobe of probable seizure origin. Raw t-maps from SPM analysis of the same data were used as input for automated analysis (AA). Cortical masks representing each lobe were created from the AAL brain template (Tzourio-Mazoyer, 2001). A custom MATLAB (MathWorks, 7.14) script was then used to determine (a) proportion of activated voxels in a lobe and their (b) median and (c) mean value with five different thresholds (with uncorrected p=0.001, p=0.005, p=0.01, p=0.05, p=0.1) and in hypo- and hyperperfusion maps.Results: Highest concordance of resection site and results of automated analysis was 76% for median voxel value (with threshold p = 0.01), 67% for mean voxel value was (with threshold p=0.05), and 57% for proportion of activated voxels (same for thresholds p=0.005 and p = 0.001). Analysis of hypoperfusion changes gave poor results (highest concordance was 24%). Total computation time was approximately 15 minutes, with human interaction needed for 2 minutes. The concordance of resection site and reviewers consensus was 71% in SPM-SPECT and 48% in SISCOM. There was no significant difference in injection times or seizure duration between concordant and non-concordant group from the automated analysis (p=0.41 and p=0.32, two-sample t-test). Comparison of automated analysis with SISCOM and SPM-SPECT methods showed no significant difference in localization (SPM-SPECT p = 1; SISCOM p = 0.077, McNemar s test).Conclusions: Automated analysis of SPM-SPECT in MRI-negative temporal lobe epilepsy provides similar results to those made by consensus of blinded reviewers. The main benefits include markedly reduced time to obtain final results and lack of interobserver disagreement and bias introduced by results of other diagnostic modalities