Clustering of Medical Images - Medical Image Processing: Advanced Fuzzy Set Theoretic Techniques - page 166

to introduce bias in consequence of the large inter-user variability, which is
generally observed when manual labelling is performed and (4) it is time
consuming especially for large volumes.
In unsupervised clustering, no user's definition of training samples is
required and multidimensional data are clearly exploited.
A MATLAB ® program for fuzzy clustering is given.
An image is clustered into four groups, and pixel is used as feature. An
FCM clustering algorithm is used.
clear all;
image2=imread('tumor_2.jpg');imagergb=rgb2gray(image2);
dim=75;
image=imcrop(imagergb,[1 1 dim-1 dim-1]);
im1=double(image);
imf = im1(:);
[centerL,U,obj_fcn]=fcm(imf,4); % U = membership matrix,
center = cluster center, obj_fcn is the objective function,
4 is no. of clusters
maxU = max(U);
index1 = find(U(1,:)==maxU);
index2 = find(U(2,:)==maxU);
index3 = find(U(3,:)==maxU);
index4 = find(U(4,:)==maxU);
imf(index1)=0.5*ones;
imf(index2)=zeros;
imf(index3)=ones;
imf(index4)=0.7*ones;
imff=reshape(imf,[dim dim]);
figure,imshow (imff,[]);
7.10 Su m m a r y
This chapter presents in detail FCM clustering and hierarchical clustering.
Kernel clustering is also discussed as it is robust to noise. The kernel function
uses Gaussian or hypertangent or the radial function and these functions
follow the exponential function that is robust to noise. Clustering medical
images using kernel-based FCM and IFCM clustering is discussed in this
chapter. Clustering of tumours/clots of CT scan images of the brain is also
carried out. Intuitionistic fuzzy kernel c means clustering performs better on
noisy images.