A sparse representation-based method for parcellation of the resting brain and its application to treatment-resistant...

Parcellating brain regions into functionally homogeneous subdivisions is critical for understanding normal and abnormal brain functions.

In this study, we developed a new sparse representation-based parcellation method for functional magnetic resonance imaging (fMRI) data, and applied the new method to investigate functional insular subdivisions in treatment-resistant major depressive disorder (MDD). Realistic simulations were implemented to demonstrate the feasibility of the method. Subsequently, the method was used to parcellate the insula in a sample of fifty-six MDD patients and thirty-six healthy volunteers (HVs). The optimal number of clusters was determined by an independent test-retest dataset. Finally, differences of the functional connectivity profiles of each insular subdivision between patients and HVs were inspected.

The results from both simulated and test-retest fMRI datasets demonstrated the feasibility of the proposed elastic net-based (EN) method. With the proposed method, the insula was parcellated into four subdivisions (dorsal anterior dAI; ventral anterior vAI; middle, MI and posterior, PI). Whereas patients showed hypo-connectivity between vAI and right medial temporal lobe, there were no functional volumetric differences in insular subdivisions between MDD patients and HVs.

Results from both simulated and real fMRI datasets showed that the proposed EN method achieved higher accuracy than least absolute shrinkage and selection operator-based (LASSO) method.

These findings suggest that EN-based parcellation has the potential to be a useful addition to the parcellation techniques for fMRI data, and provide evidence of decreased functional connectivity without functional volumetric changes of the insula in treatment-resistant MDD.