MRI Brain Image Segmentation Using Interactive Multiobjective Evolutionary Approach

MRI Brain Image Segmentation Using Interactive Multiobjective Evolutionary Approach

Anirban Mukhopadhyay (University of Kalyani, India)
DOI: 10.4018/978-1-5225-0058-2.ch002


The problem of image segmentation is frequently modeled as a problem of clustering the pixels of the images based on their intensity levels. In some recent studies, multiobjective clustering algorithms, where multiple cluster validity measures are optimized simultaneously for yielding robust clustering solutions have been proposed. It has been observed that the same set of validity measures optimized simultaneously do not generally perform well for all image datasets. In view of this, in this article, an interactive approach for multiobjective clustering is proposed for segmentation of multispectral Magnetic Resonance Image (MRI) of the human brain. In this approach, a human decision maker interacts with the multiobjective evolutionary clustering technique during execution in order to obtain the final clustering, the suitable set of validity measures for the input image, as well as the number of clusters by employing a variable-length encoding of the chromosomes. The effectiveness of the proposed method is demonstrated on many simulated normal and MS lesion MRI brain images.
Chapter Preview


Image segmentation refers to the procedure of separating the pixels of an image into multiple non-overlapping, homogeneous and meaningful regions (Gonzalez, 1992). These regions are generally strongly associated with the objects in the image. Segmentation plays a vital role for analysis of medical images in computer-aided diagnosis and therapy. Automatic segmentation of MRI brain images into different tissue classes, such as gray matter (GM), white matter (WM) and cerebrospinal fluid (CSF) is very important in clinical study and neurological pathology. Further, the quantization of GM and WM region volumes may be of major interest in understanding different neurological disorders and diseases.

The problem of segmentation of MRI brain images in different tissue classes is often posed as clustering the image pixels in the intensity space (Mukhopadhyay, 2009a, 2011). Clustering (Jain,1988) is a popular unsupervised pattern classification technique that partitions a set of 978-1-5225-0058-2.ch002.m01 objects into 978-1-5225-0058-2.ch002.m02 groups based on some similarity/dissimilarity metric where the value of 978-1-5225-0058-2.ch002.m03 may or may not be known a priori. Unlike hard clustering, a fuzzy clustering algorithm produces a $K \times n$ membership matrix 978-1-5225-0058-2.ch002.m04, 978-1-5225-0058-2.ch002.m05 and 978-1-5225-0058-2.ch002.m06, where 978-1-5225-0058-2.ch002.m07 denotes the membership degree of pattern 978-1-5225-0058-2.ch002.m08 to cluster 978-1-5225-0058-2.ch002.m09. For probabilistic non-degenerate clustering 978-1-5225-0058-2.ch002.m10 and978-1-5225-0058-2.ch002.m11, 978-1-5225-0058-2.ch002.m12 (Bezdek, 1981). Due to the inherent noisy nature of MRI images, fuzzy clustering is more appropriate in segmentation of MRI imagery.

Complete Chapter List

Search this Book: