Non-standard templates for non-standard populations: optimizing template selection for voxel-based morphometry pre-processing

Abstract

The human brain is a complex and powerful organ, directing every aspect of life from somatosensory and motor function to visceral responses to higher order cognition. Neurological and psychiatric disorders often disrupt normal functioning. While the clinical symptoms of such disorders are known, their biological underpinnings are not as clearly characterized. Structural neuroimaging is a powerful, non-invasive tool that can play a critical role in finding biomarkers of these illnesses. Currently, variations in pre-processing techniques yield inconsistent and conflicting results. As neuroimaging is a nascent branch of medical research, gold standards in imaging methodologies have not yet been established. Quantitatively validating and optimizing the way these images are preprocessed is the first step towards standardization. Voxel-based morphometry (VBM) is one technique that is commonly used to compare whole-brain structural differences between groups. Statistical tests are used to compare intensities of voxels throughout all brain scans in each group. In order to ensure that comparable voxels are being tested, the images must be fitted into a common space, which is done through image preprocessing. Spatial normalization to templates is an early pre-processing step that is executed unreliably as many options for both templates and normalization algorithms exist. To determine the effect variations in template usage may cause, we utilized a VBM approach to detect simulated lesions. Template performance was analyzed by comparing the accuracy with which the lesion was detected.