MULTI-MODAL AND MULTI-REGION DISTANCE MODEL FOR NEUROIMAGING: APPLICATION TO ABCD STUDY.

Area of Science:

• Neuroimaging
• Biostatistics
• Data Science

Background:

• Large-scale neuroimaging studies integrate multiple data types (e.g., fMRI, dMRI, sMRI) for comprehensive brain analysis.
• Current joint inference methods struggle with complex cross-modality covariance, limiting interpretability and statistical power.
• Existing distance-based ANOVA extensions lack interpretable parameters and require intensive computation.

Purpose of the Study:

• To introduce a novel semiparametric framework, U-statistics-based Generalized Estimating Equation (UGEE), for multimodal neuroimaging data analysis.
• To develop a method that flexibly models cross-modality covariance and provides interpretable regression coefficients.
• To offer a computationally efficient alternative to existing permutation-based inference methods.

Main Methods:

• Developed a UGEE framework unifying univariate and multivariate distance models by regressing pairwise dissimilarities on covariates.
• Utilized efficient influence functions for asymptotic efficiency, robustness, and scalability.
• Evaluated the method via simulations and analysis of the Adolescent Brain Cognitive Development dataset.

Main Results:

• The UGEE framework accurately estimates modality, group, and interaction effects in multimodal neuroimaging data.
• The method effectively disentangles location and scale effects and quantifies inter-modality differences.
• Achieved a 100-fold speed-up in computation compared to permutation-based approaches.

Conclusions:

• The proposed UGEE framework offers a general, computationally efficient tool for semiparametric inference on multimodal data.
• This method is particularly suitable for large-scale neuroimaging applications, enhancing statistical efficiency and interpretability.
• UGEE provides a robust and scalable approach for integrated brain-phenotype association studies.