Unsupervised segmentation of brain tissues using multiphase level sets on multiple MRI sequences

Automated segmentation of brain tissues is usually performed on T1-weighted magnetic resonance images, due to the high resolution that characterizes them. The major problems related to this sequence are caused by partial volume effects and bias field inhomogeneity. Moreover, when dealing with neurodegenerative diseases, the presence of lesions, either in gray-matter or white-matter, may cause incorrect segmentation or tissue classification; this affects in turn the correct identification of lesion type and position, making an automatic evaluation of lesion burden hard. In order to selectively suppress the contribution of specific tissues, several MR sequences have been designed. In particular, in this work we use two different MRI sequences, FLAIR and DIR, to segment gray matter, white matter and cerebro-spinal fluid adopting a multiphase level set framework. These two sequences allow highlighting respectively gray and white matter, and gray matter alone. Our method is based on the multiphase piecewise constant active contour model without edges extended to vector-valued functions, i.e. applied to multimodal MR images.