Segmentation of Brain Tissue Using Improved Kernelized Rough-Fuzzy C-Means Technique

  • Bhanu Prasad M C, Dr. Manoj Kumar


Brain magnetic resonance image (MRI) analysis is a vibrant and challenging area of research in the domain of bio-medical engineering or machine learning. Identification or detection of any kind of abnormalities such as tumor, edema, etc., at an early stage is crucial for radiologists and medical practitioners to suggest and employ appropriate treatment strategy. Aside from the fact that MRI is one of the most widely-used diagnostic imaging techniques, it has additional advantages such as the ability to produce detailed multimodal scans of different tissues of the human brain with a variety of contrast using the three magnetic resonance modalities (T1 weighted, T2 weighted, and T3 weighted) that can be used to detect abnormal changes in various tissues and organs. In contrast, Brain MRI suffers from the possibility of noise, outliers, and other aberrations. Normally, standard unsupervised machine learning techniques often fail to produce accurate segmentation results due to the possible presence of noise and outliers in brain MRI. Moreover, various brain tissue regions such as white matter (WM), gray matter (GM), and cerebrospinal fluid (CSF), etc., are usually imprecise, overlapping, vague, unclear and difficult to understand in its basic form. The separability of distinct tissues is also not linear. In view of the aforesaid issues, effort is made in the present thesis to address the following challenges such as (i) presence of noise in the brain MRI datasets, (ii) vague, uncertain, overlapping and indiscernible complex brain tissue regions, (iii) non-linearly separable different brain tissue regions and (iv) inaccurate performance of the traditional unsupervised methods. In this respect, the present thesis proposes some robust and novel unsupervised, semi-supervised, and ensemble learning techniques by judicious amalgamation of (a) The use of a kernel technique increases the likelihood of distinct brain tissue areas being linearly separable in its original feature space even if they would not have been otherwise. To cope with probable noise, outliers, and other artefacts, we use spatial contextual information (b), a fuzzy set (c), a rough set (d), and finally a fuzzy set, and (e) semi-supervision to guide the clustering process in better direction by supplying some labelled pixels with constraint seeded policy, which guides the clustering process in a better direction. Moreover, ensemble of different unsupervised algorithms is also done to have a consensus decision.

How to Cite
Bhanu Prasad M C, Dr. Manoj Kumar. (2022). Segmentation of Brain Tissue Using Improved Kernelized Rough-Fuzzy C-Means Technique. Design Engineering, (1), 4146 - 4164. Retrieved from