A multi-stage clustering approach for cerebrospinal fluid image segmentation

Abstract

Analysis of the cerebro spinal fluid (CSF) flow within brain has become increasingly important to diagnose a number of neuro degenerative disorders. Magnetic resonance imaging (MRI) is utilised to measure the CSF volumetric change in patients. However, the quality of the images is often hampered by partial volume effect, which blurred the boundary between the brain tissues and the CSF. Consequently, the accuracy of CSF analysis is reduced significantly. In this paper, we introduce a new multi-stage clustering approach to overcome this limitation. Firstly, the T1-weigthed images are fused with the corresponding T2-weigthed images. Next, the resulting images are subjected to partial volume estimation using Gaussian mixture model. The model produced by these images is later used as input in a spatial fuzzy clustering algorithm to segment the CSF flow from the brain tissues. Benchmark images obtained from Brain Web are used to validate the performance of the proposed approach. In addition, we also presented the performance of the proposed method using real MRI images taken from a number of Alzheimer’s disease patients, which evidently showed the effectiveness of the method in quantifying the CSF flow within the brain.