IC-P3-171: Scan-rescan reliability of semi-automated brain region extraction (SABRE)

Abstract

Background: The clinical applicability of longitudinal MRI studies requires validation of the reliability of the scan acquisition and tools used in the generation of data. A scan-rescan analysis using repeated scans with short interscan time intervals is important to accurately measure the reliability of the imaging data being acquired, and increase confidence in the results. Regional tissue classification techniques have potential to elucidate brain-behaviour relationships in both longitudinal and cross-sectional research. Our automated, T1-based segmentation protocol provides gray matter, white matter, and cerebrospinal-fluid (CSF) whole brain volumes after “head from brain” removal (Kovacevic et al.,2002). Semi-Automated Brain Region Extraction (SABRE) parcellates the brain into 26 regions of interest (ROI’s) (13 bilaterally), based on 15 manually identified landmarks (Dade et al.,2004). The output gives individual compartment volumes for gray, white, CSF and ventricular CSF for each of the 26 ROI’s. Because this method reliably captures all subdural/sulcal CSF, it is an ideal tool for measuring individualized and group atrophic changes regionally and globally while, unlike most automated methods, still respecting individual anatomical variability. Methods: Twenty-one participants, ranging in age from 22 to 91 years were scanned on a 1.5T GE signa scanner on two occasions with a mean interval of 11 days, using standardized 3D T1weighted, T2-weighted, and Proton Density acquisitions at each session. Two operators, blinded to subject identity and scan order performed the procedure on each subject. Results: Shrout & Fleiss Model-3 Intraclass Correlation Coefficients (ICC) were generated for each of the 26 ROI’s, comparing the two different scanning sessions. The mean ICC for all SABRE regions was 0.94, ranging from 0.83 (right anterior basal ganglia and thalamus) to 0.98, with 22 of 26 regions greater than 0.90, indicating highly reproducible results (Shrout & Fleiss,1979). Conclusions: The SABRE method appears highly reproducible giving confidence that it could be used in longitudinal studies. For hypothesis-driven brainbehaviour studies, this method offers a viable alternative between fullyautomated techniques, which are fast but do not account for individual variability and frequently do not include subdural CSF spaces, and fullymanual techniques which are accurate but time consuming, require extensive training, and are less reproducible. IC-P3-172 BRAIN MORPHOLOGY IN ELDERLY AFRICANAMERICANS, CARIBBEAN HISPANICS, AND CAUCASIANS FROM NORTHERN MANHATTAN