Bayesian Hierarchical Regression for Robust Multi-Site Neuroimaging Analysis

Multi-site neuroimaging studies offer increased statistical power and generalizability but are often plagued by sitespecific variations that can confound analyses and lead to biased conclusions. To address these challenges, we propose a novel Bayesian hierarchical regression model that explicitly accounts for site-specific effects and heteroscedastic variance. Our approach incorporates site as a factor in the model, allowing for the estimation of site-specific effects while simultaneously modeling heteroscedasticity to capture varying levels of variability across sites. This not only improves the accuracy of the analysis but also provides a more nuanced understanding of the data. We demonstrate the effectiveness of our method on a large multisite neuroimaging dataset, showing improved performance in regression and anomaly detection tasks compared to conventional methods. Our findings highlight the importance of explicitly accounting for site-specific effects and heteroscedasticity in multisite neuroimaging studies. The proposed Bayesian hierarchical regression model offers a robust and flexible framework for analyzing multi-site neuroimaging data, leading to more reliable and generalizable conclusions.