Abstract
1694
Objectives: Statistical comparison of functional brain volumes often requires an alignment step of each brain volume into a common template space to ensure that homologous structures are in the same location. The aim of the study was to compare the ability of the Statistical Parametric Mapping (SPM) and MIMneuro software packages to reduce anatomical variability when performed directly on PET brain volumes. Methods: T1-weighted MRI scans were obtained for 10 subjects. Four subjects also received 11C-DTBZ PET scans and six received FDG-PET scans, and each were registered to a template volume using both SPM and MIMneuro. Two metrics were utilized to quantify registration accuracy: 1) correlation between the template volume and the registered subject volume and 2) area under ROC curves (aROCs) for volumes of interest (VOIs). VOIs were defined in template space which ranged from 100% specific to 100% sensitive for each of 13 anatomical structures as manually defined on the co-registered MRI volumes for each of the ten subjects. Correlation is inversely proportional to functional (and anatomical) variability while the aROC metric is inversely proportional to anatomical variability. Results: The mean correlation was 0.869 for SPM and 0.894 for MIMneuro. The difference was significant using the student’s t-test at a p-value of 0.015. The difference is also visually apparent, especially in the basal ganglia. The mean aROC was 0.860 for SPM and 0.869 for MIMneuro. While the aROCs were very similar for most of the structures, SPM showed greater consistency in registering the basis pontis and cerebellar vermis while MIMneuro showed greater consistency in registering the caudate, cerebellar hemisphere and the retrosplenial area. Conclusions: Both algorithms performed similarly at the task of reducing anatomic variation in the 13 structures analyzed. The significantly better correlation results for MIMneuro in comparison to SPM may indicate better performance in regions other than the 13 structures compared with the aROC metric. Additional aROC comparisons should be made to determine whether this is the case. Both algorithms appear adequate for spatial normalization of PET brain volumes.
- Society of Nuclear Medicine, Inc.