"Voxel-based morphometry" should not be used with imperfectly registered images

Area of Science:

• Neuroimaging
• Statistical analysis
• Brain morphology

Background:

• Introduced in 2000, voxel-based morphometry (VBM) by Ashburner and Friston offers a standardized method for comparing gray matter concentrations between subject groups.
• The VBM method involves image segmentation, spatial normalization, smoothing, and voxelwise statistical analysis using Gaussian random fields.

Discussion:

• This commentary highlights a critical interaction between VBM's spatial normalization and voxelwise comparison steps.
• This interaction can introduce quantitative confounds, compromising the reliability of group difference inferences.
• The accuracy of VBM is questioned in regions where the spatial normalization algorithm struggles with image gradients.

Key Insights:

• VBM analysis is unreliable in areas with poor spatial normalization, specifically where image gradients are not robustly registered.
• The statistical inferences derived from VBM are uninformative about true group differences in these problematic regions.
• The validity of the Ashburner and Friston VBM method is limited to areas with clear image gradients.

Outlook:

• Further research is needed to refine VBM algorithms to mitigate confounds introduced by spatial normalization.
• Development of more robust spatial normalization techniques is crucial for accurate VBM analysis.
• Alternative or complementary morphometric methods may be required for comprehensive brain structure analysis.