Stratified voxel-based morphometry (sVBM)

摘要: In neuroimaging, voxel based morphometry (VBM) has been a valuable tool of identifying brain-wide differences between populations. One of the key elements of VBM is to define a space for voxelwise comparisons. However, errors in this mapping to common space and variations of brain morphology, both natural and pathologic can result in false positives. In this work we explore a new framework, where a spatially varying morphological similarity graph is created between pairs of images. This graph is then used to stratify natural and pathological variability in a VBM-like setting. In contrast to VBM, which describes the group differences on an average brain morphology, sVBM describes how different brain morphologies are independently affected by pathology. Due to its pairwise nature, this technique provides smoother and better localised differences between populations, possibly providing novel insights into the homogeneity of pathological effects for different brain morphologies.