Spatial normalization for voxel-based lesion symptom mapping: impact of registration approaches

Area of Science:

• Neuroimaging
• Neurology
• Medical Image Analysis

Background:

• Voxel-based lesion symptom mapping (VLSM) links lesion location to clinical outcomes.
• Spatial normalization is a critical pre-processing step in VLSM.
• Optimal registration methods for VLSM remain debated, impacting statistical analysis.

Purpose of the Study:

• To evaluate four common spatial normalization registration approaches for VLSM.
• To assess the downstream effects of these registration methods on VLSM statistics.
• To determine the most suitable registration method for stroke imaging data.

Main Methods:

• Compared affine (AR), nonlinear (NLR), cost function masking (CFM), and enantiomorphic (ENR) registration.
• Utilized FLAIR MRI data from 122 acute ischemic stroke patients.
• Assessed lesion-voxel relations to 30-day modified Rankin Scale (mRS) using univariate and multivariate statistics.

Main Results:

• Key brain areas for mRS outcomes were consistent across registration methods.
• Nonlinear methods (NLR, CFM, ENR) introduced image distortions.
• Nonlinear registration resulted in insufficient alignment of local structures and adjacent brain areas.

Conclusions:

• The benefits of nonlinear registration variants for spatial normalization in VLSM appear questionable for stroke studies.
• Image distortions from nonlinear registration introduce uncertainties in VLSM analyses.
• Affine registration may be a more reliable choice for spatial normalization in similar VLSM studies.