Estimation of brain functional connectivity (FC) is essential for understanding the functional organization in the brain and for identifying changes occurring due to neurological disorders, development, treatment, and other phenomena. Independent component analysis (ICA) is a matrix decomposition method that has been used extensively for estimation of brain functional networks and their FC. However, estimation of FC via ICA is often sub-optimal due to the use of ad-hoc methods or need for temporal dimension reduction prior to traditional ICA methods. Bayesian ICA methods can avoid dimension reduction, produce more accurate estimates, and facilitate inference via posterior distributions on the model parameters. In this paper, we propose a novel, computationally efficient Bayesian ICA method with population-derived priors on both the temporal covariance, representing FC, and the spatial components of the model. We propose two algorithms for parameter estimation: a Bayesian Expectation-Maximization algorithm with a Gibbs sampler at the E-step, and a more computationally efficient variational Bayes algorithm. Through extensive simulation studies using realistic fMRI data generation mechanisms, we evaluate the performance of the proposed methods and compare them with existing approaches. Finally, we perform a comprehensive evaluation of the proposed methods using fMRI data from over 400 healthy adults in the Human Connectome Project. Our analyses demonstrate that the proposed Bayesian ICA methods produce highly accurate measures of functional connectivity and spatial brain features. The proposed framework is computationally efficient and applicable to single-subject analysis, making it potentially clinically viable.
翻译:暂无翻译