Abstracts

Rationale: Single photon emission computed tomography (SPECT) is a radioisotope-based imaging modality that can demonstrate seizure foci based on ictal perfusion changes. The diagnostic yield of ictal SPECT can be augmented through the use of statistical parametric mapping (SPM). Here we demonstrate and validate a means of automated localization of SPM-SPECT abnormalities using a modified version of the automated anatomic labeling (AAL) atlas. Methods: 29 patients who had undergone ictal SPECT with subsequent SPM analysis and who subsequently received resective epilepsy surgery were identified, of whom 14 had undergone SPECT studies that were considered at surgical epilepsy conference to have had possible localizing value. The localization of the area of resection was identified using postoperative MRI where available (7/14) or by means of chart review. The lobar localization of the SPM-SPECT hyperperfusion abnormality was obtained automatically using three different modifications of the AAL atlas:  1) a version of the AAL atlas from which subcortical regions were removed; 2) a version of the AAL atlas in which gray matter (GM) ROIs were expanded to cover nearby white matter (WM) regions to accommodate activations that appear in WM regions due to the low resolution of SPECT and smoothing in the processing, following which process subcortical ROIs were removed, and 3) a version of the atlas in which in addition to the expansion of GM ROIs and removal of subcortical ROIs, occipital ROIs were also removed to compensate for the occipital hyperperfusion seen in images of patients whose eyes are open at the time of tracer injection. Localization of the SPECT hyperperfusion abnormalities were determined both by the atlas ROI containing the greatest mean SPM SPECT intensity and the ROI containing the most intense single voxel. Results: Among the 14 post-resection patients who had been considered to have localizing SPECT studies, automated localization of SPM-SPECT hyperperfusion using the AAL atlas without subcortical regions was concordant with the region of resection in 9/14 cases, 6 of whom were seizure-free at last follow-up (median follow-up = 258 days). In this case there was no difference between using average intensity over an ROI or max voxel intensity to identify the relevant AAL ROI. When using an atlas with expanded GM ROIs and with subcortical regions excluded, ROIs as localized by average-intensity and max-voxel methods were concordant with the region of resection in 11/14 and 9/14 cases, respectively. With use of an atlas with expanded GM ROIs and with subcortical and occipital regions excluded, ROIs as localized by average-intensity and max-voxel methods were concordant with the region of resection in 12/14 and 10/14 cases, respectively. Each of the methods described localized SPM-SPECT intensity to the resection site in 6/7 of the cases with both localizing SPM-SPECT and postoperative MRI. Conclusions: These results provide support for the use of automated localization techniques for SPM-SPECT. Such an approach may have a role in the multimodality evaluation of surgical epilepsy patients. Funding: None