Segmentation of brain tissue using improved kernelized rough-fuzzy c-means technique

Abstract

Brain magnetic resonance imaging (MRI) data is a hot topic in the domains of biomedical engineering and machine learning. Without locating anomalies, such as tumors and edema, radiologists and other medical experts cannot effectively recommend or administer therapy for patients. Having three different magnetic resonance techniques (T1 weighted, T2 weighted, and T3 weighted), MRI can produce detailed multimodal scans of different human brain tissues with varying contrast, which can help pinpoint the source of any abnormalities. The cerebrospinal fluid (CSF), white matter (WM), and grey matter (GM) are all components of the brain, and their boundaries are sometimes hazy and difficult to nail down. In light of the problems above, this paper makes an effort to tackle issues like: i) the noise that exists in the brain datasets for MRI, ii) the fuzziness, uncertainty, overlap, indiscernibility of complex brain tissue regions, iii) the inability of traditional unsupervised methods to reliably distinguish between various brain tissue locations, and iv) ineffective performance. We propose some robust techniques by utilise spatial contextual data, a rough set, a fuzzy set, and ultimately a fuzzy set to steer the clustering process in a better direction, allowing it to deal with likely noise, outliers, and other artifacts.