Radial basis function-sparse partial least squares for application to brain imaging data.

Area of Science:

• Neuroimaging
• Statistical modeling
• Medical research

Background:

• Magnetic resonance imaging (MRI) is crucial for brain morphology research.
• Analyzing high-dimensional MRI data alongside clinical variables presents computational challenges.
• Existing methods often require prespecified brain regions, limiting discovery.

Purpose of the Study:

• To introduce a novel statistical method, radial basis function-sparse partial least squares (RBF-sPLS), for brain morphology analysis.
• To investigate the relationship between clinical characteristics and brain morphology using 3D MRI data.
• To overcome computational difficulties in high-dimensional neuroimaging data analysis.

Main Methods:

• Application of radial basis function-sparse partial least squares (RBF-sPLS) to 3D MRI data and 73 clinical variables.
• Simultaneous selection of effective brain regions and clinical characteristics using sparse modeling.
• Dimensionality reduction to handle high-dimensional neuroimaging data.

Main Results:

• RBF-sPLS identified the temporal lobe associated with aging and the occipital lobe associated with anemia in chronic kidney disease patients.
• A simulation study confirmed the high accuracy of RBF-sPLS in extracting relevant brain regions.
• The method effectively handles potential correlations within voxel data and clinical variables.

Conclusions:

• RBF-sPLS is a powerful and accurate tool for exploring brain morphology and its relationship with clinical factors in neuroimaging research.
• The method facilitates the discovery of novel associations between specific brain regions and diseases or conditions.
• RBF-sPLS offers a significant advancement over existing methods by enabling simultaneous feature selection and reducing computational burden.