Brain effective connectome based on fMRI and DTI data: Bayesian causal learning and assessment

Area of Science:

• Neuroscience
• Computational Neuroscience
• Brain Imaging Analysis

Background:

• Current methods for discovering the brain's Effective Connectome (EC) are limited by fMRI data's short sample size, poor temporal resolution, and the high dimensionality of brain connectomes.
• Existing causal discovery methods struggle to accurately identify ECs using only fMRI data, hindering a comprehensive understanding of brain organization.

Purpose of the Study:

• To introduce novel Bayesian causal discovery frameworks (Bayesian GOLEM and Bayesian FGES) that leverage Diffusion Tensor Imaging (DTI) data as prior knowledge to enhance EC discovery.
• To develop and validate a new computational metric, the Pseudo False Discovery Rate (PFDR), for numerically assessing the accuracy of EC discovery in neuroscientific studies.
• To evaluate the accuracy and reliability of the proposed Bayesian methods and the PFDR metric using synthetic, hybrid, and empirical data.

Main Methods:

• Development of two Bayesian causal discovery frameworks: Bayesian GOLEM (BGOLEM) and Bayesian FGES (BFGES).
• Integration of Diffusion Tensor Imaging (DTI) data as prior information within the Bayesian frameworks.
• Introduction and application of the Pseudo False Discovery Rate (PFDR) as a novel accuracy metric for causal discovery.
• Validation using simulation studies on synthetic and hybrid datasets (Human Connectome Project DTI + synthetic fMRI) and empirical data from the Human Connectome Project (HCP).
• Assessment of reproducibility using the Rogers-Tanimoto index on test-retest data.

Main Results:

• The proposed Bayesian methods (BGOLEM and BFGES) significantly improve the accuracy and reliability of Effective Connectome (EC) discovery compared to traditional methods.
• The Pseudo False Discovery Rate (PFDR) metric effectively quantifies the accuracy of EC discovery, demonstrating the superiority of the Bayesian approaches on HCP data.
• The Bayesian methods yield more reproducible ECs, as evidenced by higher Rogers-Tanimoto index scores on test-retest data.

Conclusions:

• The novel Bayesian causal discovery frameworks offer a significant advancement in accurately and reliably identifying the brain's Effective Connectome (EC).
• The developed PFDR metric provides a reliable computational tool for assessing causal discovery accuracy in neuroimaging.
• These advancements hold substantial potential for deepening the understanding of brain functionality and organization through improved connectome mapping.