SEGMENTING THE VASCULARIZATION OF THE CANCEROUS LIVER TISSUE IMAGE

Abstract

In this paper, it is to segment the vascularization of the liver tissue image for automated hepatocarcinoma differentiation. From the pathological significance, the area of vascularization is one of the improtant parameter for pathological analysis. The methods employed for vascularization segmentation are Otsu's method, Otsu_2D method, fuzzy c-means, priccipal component transformation/median-cut, and evolutionary autonomous agents. In the automatic diagnosis, three parameters, i.e. fractal dimension, nuclear counts, and vascularization area, are used the input feature vector for the neural networks. Two types of neural network used in the diagnosis stage are learning vector quantization and probabilistic neural network. From the results, it is shown that PCT/Median_ Cut combined with VLQ would obtain the best results for accurancy about 90%. Finally, a visual programming workspace is also developed for easy to use.