A powerful and efficient multivariate approach for voxel-level connectome-wide association studies

Area of Science:

• Neuroimaging analysis
• Computational neuroscience
• Statistical modeling

Background:

• Resting-state functional magnetic resonance imaging (rsfMRI) is crucial for understanding brain function.
• Voxel-level connectomes offer detailed insights into brain connectivity patterns.
• Existing multivariate analysis methods face limitations in identifying complex voxel-phenotype associations.

Purpose of the Study:

• To introduce structured kernel principal component regression (sKPCR), a novel multivariate method for voxel-level connectome analysis.
• To enhance the identification of linear and non-linear voxel-phenotype associations using rsfMRI data.
• To improve computational efficiency and statistical power compared to existing methods.

Main Methods:

• sKPCR employs structured kernel principal component analysis for signal extraction from connectivities.
• An adaptive regression model is used for testing voxel-phenotype associations.
• The method models spatial data structure during dimension reduction and adaptively selects dimensions for testing.

Main Results:

• Simulations demonstrate sKPCR's ability to control false-positive rates and its superior power over GLM, MDMR, aSPU, and LSKM.
• sKPCR exhibits faster computation speeds by reducing the cost of permutation tests.
• Real data analysis in schizophrenia datasets showed improved between-sites reproducibility and overlap with meta-analysis findings.

Conclusions:

• sKPCR is a powerful and computationally efficient method for identifying voxel-phenotype associations in rsfMRI connectome data.
• The method effectively handles both linear and non-linear signals and offers advantages in statistical power and speed.
• sKPCR demonstrates significant utility in clinical neuroscience research, particularly in identifying brain differences in conditions like schizophrenia.