Multi-tissue Classification of Diffusion-Weighted Brain Images in Multiple System Atrophy Using Expectation Maximization Algorithm Initialized by Hierarchical Clustering

摘要

Multiple system atrophy (MSA) is a well-known neurodegenerative disorders that present parkinsonism syndrome and autonomic dysfunction. Patients with MSA who have the combination of parkinsonism and cerebellar ataxia are referred to as MSA-C. Brain diffusion-weighted imaging (DWI) offers the potential for objective criteria in the diagnosis of MSA. We aim to develop an automatic method to segment out the abnormal whole brain area in MSA-C patients based on the 13-direction DWI raw data. The whole brain DWI raw data of fifteen normal subjects and nine MSA-C patients were analyzed. In this study, we proposed a novel method to perform tissue segmentation directly based on the directional information of the DWI images, rather than using the parametric images, such as fractional anisotropy (FA) and apparent diffusion coefficient (ADC) as in the previous literatures. Specifically, a hierarchical clustering (HC) technique was first applied on the down-sampled data to initialize the model parameters for each tissue cluster followed by automatic segmentation using the expectation maximization (EM) algorithm. Our results demonstrate that the HC-EM is effective in multi-tissue classification, namely, the cerebrospinal fluid, gray matter, and several areas of white matters, on the DWI raw data. The segmented patterns and the corresponding intensities of thirteen directions of the cerebellum in MSA-C patients showed the decrease of the anisotropy, which were evidently different from the results in normal subjects.