MULTIVARIATE HYPOTHESIS TESTING OF DTI DATA FOR TISSUE CLUSTERING

In this work we investigate the feasibility and effectiveness of unsupervised tissue clustering and classification algorithms for DTI data. Tissue clustering and classification are among the most challenging tasks in DT image analysis. While clustering separates acquired data into objects, tissue classification provides in-depth information about each region of interest. The unsupervised clustering algorithm utilizes a framework proposed by Hext and Snedecor, where the null hypothesis of diffusion tensors arising from the same distribution is determined by an F-test. Tissue type is classified according to one of three possible diffusion models (general anisotropic, prolate, or oblate), which is determined with a parsimonious model selection framework. This approach, also adapted from Snedecor, chooses among different models of diffusion within a voxel using a series of F-tests. Both numerical phantoms and DWI data obtained from excised rat spinal cord are used to test and validate these tissue clustering and classification approaches.