|
Abstract:
|
Brain connectomics using neuroimaging has become increasing important to advance understanding of neural circuitry among healthy as well as diseased human brains. There are major challenges in brain connectome research including the high dimensionality of brain networks, unknown neural circuits underlying the observed connectivity, and the large number of brain connections leading to spurious findings. We present statistical methods that aim to improve the reliability and reproducibility in brain connectome research. The methods provide fully data-driven decomposition of observed network measures to reveal underlying neural circuits. We tackle the challenges in brain connectiomics with statistical strategies such as low-rank factorization, novel angle-based sparsity regularization and an automatic method for adaptively selecting the tuning parameters. We propose a computationally efficient iterative Node-Rotation algorithm to solve the non-convex optimization problems. By applying the methods to large-scale neuroimaging studies, we obtain more reliable findings and discover biologically insightful connectivity traits that are not revealed with the existing method.
|
