To evaluate the newly developed diffeormorphic image registration framework using stationary velocity fields parameterized by wendland kernel bundle framework in atrophy estimation. In this study, we compare the diagnostic group separation (Alzheimer's and Normals) abilities of the proposed framework against other state-of-art registration schemes and the Boundary shift integral (BSI) based on atrophy scores in several brain regions.
METHOD AND MATERIALSBaseline and month 12 MRI scans from the "complete annual year 2 visits" 1.5-T standardized ADNI dataset were used [169 normal controls (NC), 101 AD]. Segmentations for atrophy quantifications were obtained using Freesurfer cross-sectional pipeline. Each image was corrected for intensity inhomogenities using N3 from freesurfer. Each of baseline and month 12 scans were non-linearly aligned using the proposed framework and existing methods like SyN, NiftyReg, LCC-Demons. Atrophy was then estimated from the deformation field of the proposed framework using the proprietary Cube Propagation and on the rest, using Jacobian determinants. Atrophy was estimated in the regions of whole brain (WB), hippocampus (Hip), Ventricles, Medial temporal lobe (MTL), Cortical gray matter (CGM), entorhinal cortex (ENCTX) and fusiform gyrus (FG). BSI was also used to evaluate atrophy in the regions of WB, Hip and Ventricles.
RESULTSThe proposed framework yields better AUC and Cohens'D for AD v/s NC when compared to the other registration schemes. The highest separation (AUC/Cohen's D) among the registration frameworks was using the proposed framework - WB 0.76/ 0.94, hippocampus 0.82/1.26, MTL 0.86/1.43, CGM 0.85/1.29, ENCTX 0.80/1.13 and FG 0.76/0.98. Overall, BSI provided a better separation on WB (0.81/1.18), hippocampus (0.86/1.15). However, BSI was not designed to provide scores for any other region.
CONCLUSIONAlthough, BSI provides a better separation, the method can be used only in regions the software is designed for, for instance whole brain, hippocampus and ventricles. The proposed registration framework not only provides good comparable group separation (and better than other registration frameworks), it provides the flexibility to measure atrophy in any user-defined region.
CLINICAL RELEVANCE/APPLICATIONThe proposed method can reliably estimate atrophy in any brain region unlike BSI which is specifically designed to estimate atrophy only in certain regions of the brain.