New Approaches for the Analysis of the Brain's Resting State

Resting-state functional magnetic resonance imaging (fMRI) has increasingly gained attention since its introduction fifteen years ago. Its simple data collection procedure makes it a potentially useful clinical tool to investigate reorganization or adaptation of the brain's functional connections in the presence of neurological disease. Methods of data analysis, however, are not well established, and there are several pitfalls in resting-state data processing. In this thesis, specific problems associated with region-of-interest (ROI)-based analysis methods are addressed, and new methods to overcome these problems are developed and introduced. Specifically, an algorithm for ROI selection based on its internal connectivity is proposed as a means to objectively select regions for connectivity analysis without the need for a task-based fMRI localizer. Next, a connectivity calculation is introduced that is less sensitive to image noise and artifacts; this calculation is based on a procedure that normalizes connectivity in a given brain region to that of the connectivity of the seed with the seed itself. Furthermore, a time-frequency approach based on the Stockwell transform is introduced to measure similarity between seed and target region signals, without assuming signals are stationary. This method is less sensitive to inadvertent and unwanted brain activation occurring at unpredictable times and over unpredictable frequency ranges. Finally, the proposed methods are used in a preliminary clinical application to determine resting-state connectivity in the motor network of stroke patients with a motor deficit during the acute phase and after recovery. The studies in this thesis answer some problems associated with ROI-based resting-state analysis techniques, and will help establish a framework for ROI-based analysis with higher consistency and reliability.