Unsupervised segmentation of brain tissue in multivariate MRI

In this paper, we present an unsupervised, automated technique for brain tissue segmentation based on multivariate magnetic resonance (MR) and spectroscopy images, for patients with gliomas. The algorithm uses spectroscopy data for coarse detection of the tumor region. Once the tumor area is identified, further processing is done on the FLAIR image in the neighborhood of the tumor to determine the hyper-intense abnormality in this region. Areas of contrast enhancement and necrosis are then identified by analyzing the FLAIR abnormality in gadolinium-enhanced T1-weighted images. The healthy brain tissue is then segmented into white matter, gray matter, and cerebrospinal fluid (CSF) using a hierarchical graphical model whose parameters are estimated using the EM algorithm.