|
Abstract:
|
Network analysis of fMRI data has become an important tool to understand brain organization. Among the existing methods, partial correlation has shown great promises in accurately detecting true connections. In this paper, we propose an efficient and reliable statistical method for estimating partial correlation in large-scale brain network modeling. Our method derives partial correlation from precision matrix estimated via CLIME and includes a new Dens-based selection method providing a more informative and flexible tool to select tuning parameter. Based on simulation, the Dens-based method demonstrates comparable or better performance than existing methods. We applied our method to rs-fMRI data from PNC study and show that partial correlation removed considerable between-module connections identified by correlation, suggesting these connections were likely caused by global or third party effects. We also find that the most significant direct connections are between homologous brain locations in left and right hemisphere, and the sparse regularization has more shrinkage effects on negative connections than on positive connections, supporting previous biological findings.
|
