A Bayesian spatial model for neuroimaging data based on biologically informed basis functions.

Area of Science:

• Neuroimaging analysis
• Computational neuroscience
• Brain mapping

Background:

• Voxel-based analysis is standard but lacks biological meaning.
• Image resolution dictates voxel size, limiting biological relevance.
• Need for a neuroimaging analysis method with biologically meaningful units.

Purpose of the Study:

• Propose a multivariate spatial model using multiscale basis functions.
• Characterize neuroimaging data based on functional brain anatomy.
• Demonstrate utility in predicting dopamine function and classifying disorders.

Main Methods:

• Developed a Bayesian framework for estimating data as a weighted combination of basis functions.
• Utilized generic, structural MRI-derived, and resting-state fMRI-derived basis sets.
• Applied the model to quantitative SPECT imaging of striatal dopamine and classification of parkinsonian disorders.

Main Results:

• A model with ~50 multiscale functional basis functions accurately represented striatal dopamine activity.
• Functional basis functions, particularly from Instantaneous Connectivity Parcellation (ICP), yielded the most parsimonious models.
• Resting-state fMRI-derived functions outperformed structural and generic sets for dopamine function representation.
• The ICP approach excelled in classifying parkinsonian disorders, surpassing voxel-based methods.

Conclusions:

• The proposed spatial model offers a biologically informed and efficient alternative to voxel-based neuroimaging.
• Multiscale functional basis functions derived from resting-state fMRI improve the representation of brain function.
• This framework demonstrates translational validity for clinical applications like Parkinson's disease diagnosis.