SVM based pattern recognition of microscopic liver images

In this paper microscopic liver images are considered. The aim is to provide an automatic classification of the liver tissue in order to identify abnormal regions so that these regions may be deeper investigated by the experts. The classification procedure we introduce in this paper consists of three distinct steps. The image is first segmented with respect to texture properties in order to obtain a first classification of the regions present in the data. Then appropriate features are selected and extracted in the different regions of the resulting segmented image. Finally a pattern recognition model is adopted in order to linearly separate two distinct kinds of regions in the feature space described by the set of selected original features. The proposed algorithmic procedure has the aim of automatically inferring from images a sort of digital signature characterizing different liver tissues in order to highlight specific regions of practical interest from the expert's point of view and, hence, to support the experts in medical diagnosis.