SAN

SAN harmonizes vertex-level cortical thickness data across scanners by modeling and correcting spatial autocorrelation heterogeneity to preserve homogeneous covariance for multi-site neuroimaging analyses.

Key Features:

• Preservation of Homogeneous Covariance: Ensures consistency of covariance structures in vertex-level cortical thickness data across different scanners.
• Explicit Gaussian Process Modeling: Uses an explicit Gaussian process model to characterize and separate scanner-invariant and scanner-specific variations.
• Targets Spatial Autocorrelation Heterogeneity: Specifically addresses spatial autocorrelation heterogeneity that can affect vertex-level cortical thickness harmonization beyond methods like ComBat.
• Integration with Existing Harmonization Methods: Can be combined with other harmonization techniques to complement correction of inter-scanner biases.
• Computational Feasibility: Implements computationally efficient procedures suitable for large neuroimaging datasets.

Scientific Applications:

• Multi-site Vertex-level Cortical Thickness Harmonization: Harmonizes vertex-level cortical thickness measurements across scanners and sites to reduce inter-scanner bias.
• Validation with SPINS Study Data: Demonstrated utility using cortical thickness data from the Social Processes Initiative in the Neurobiology of the Schizophrenia(s) (SPINS) study.
• Support for Reproducible Neuroimaging Analyses: Reduces spatial covariance heterogeneity to improve the reliability of downstream analyses.

Methodology:

Fits an explicit Gaussian process model to vertex-level cortical thickness data to distinguish scanner-invariant from scanner-specific variations and reconstruct spatially homogeneous covariance.