Gray Matter Surface-based Spatial Statistics in Neuroimaging Studies

Abstract

Neuroimaging provides an opportunity to gain valuable insight onto the microstructural changes in the brain associated with healthy growth and neurological disorders when conducting longitudinal or cross sectional studies. However, the data interpretation in this area need to be approached with extreme care as it poses challenges because of group bias and variance arising from various factors. The desire to better understand structural-functional relationship drives the need for robust frameworks to analyze structural and functional data, especially ones that can be generalized to novel types of neuroimaging data. This dissertation develops image analysis strategies that focus on improving statistical power in quantifying brain microarchitecture features for conducting group studies in gray matter. The gray matter cerebral cortex is less than 5 mm thick, yet is key to many brain functions. To overcome the challenges of alignment issues and partial volume effects in low-resolution images like diffusion/functional MRI, the gray matter surface based spatial statistics (GSBSS) approach was developed to perform statistical analysis of multi-modal data using gray matter surfaces. Application of this technique was shown in both diffusion and functional MRI modalities in psychosis population. The main contributions of this dissertation are (1) to show that our GSBSS approach improves the statistical power for performing group studies in neuroimaging compared to that of traditional registration methods, (2) to address source of bias and variance in group studies by correcting for inter-scanner variability effects of diffusion microstructure features and constructing unbiased feature based cortical surface template, (3) to apply deep learning techniques on cortical surfaces for improved sulcal curve labeling on large datasets.