Clustering of Medical Images - Medical Image Processing: Advanced Fuzzy Set Theoretic Techniques - page 141

7
Clustering of f Medical Images
7.1 Introduction
Segmentation is a key step towards image analysis in various image process-
ing applications such as object recognition, pattern recognition and medical
imaging. It can be defined as a grouping in a parameter space where points
are associated with different sets of values of similar intensities in different
images. So, grouping is the main step of image segmentation. This type of
segmentation is called clustering which is very important in classifying dif-
ferent patterns/structures in an image. Clustering is a technique to separate
unlabelled data into finite and discrete sets. It can be done using the fuzzy
or non-fuzzy method. Traditional non-fuzzy clustering like k -means puts
data into exactly one cluster. But for overlapped data sets where some data
may be allocated to multiple clusters, k -means clustering may not analyse
the data set clearly. To achieve better clustering, fuzzy c means clustering
is used. The first fuzzy method to segment the regions of an image is the
fuzzy c means (FCM) clustering, introduced by Bezdek et al. [2]. Clustering
may be hard c means or FCM. Hard c means is a non-fuzzy method, which
is also known as k means clustering. K means clustering partitions a collec-
tion of N vectors into k groups. It executes a sharp classification in which
each object is assigned to a class or not. Also, there is very often no sharp
boundary between clusters in many real-time images. This problem can be
alleviated by associating a membership value in the interval [0, 1] to data
in every cluster such that data that have a similarity with each cluster with
membership values near 0 signify a small similarity between the sample and
the cluster and data with membership values near 1 signify a high degree
of similarity. Medical images contain a lot of uncertainties, and there are
hardly sharp boundaries present and so fuzzy clustering may be very much
beneficial. FCM partitions the data in such a way that a data point can belong
to all groups in different membership grades where an element may have
partial membership grades in several clusters - herein lies the distinction
141