Semi-Automatic Segmentation of Fibrous Liver Tissue

This article presents a semi-automatic segmentation of the fibrous liver tissue in the in-vivo liver biopsy color images. The segmentation is performed using a tree-based classifier, with decision rules as tree leaves and binary operators (AND, OR) as tree nodes. Several image´s local characteristics have been exploited, based on the image points´ intensity levels, as well as taken from the texture analysis domain (fractal dimension, FFT, Gabor filters). Their effectiveness concerning quality of extraction has been compared using real clinical images with a manual delimitation given by physicians, as a reference. A user friendly application has been developed which enables the operator to interactively create and store the classifiers. It also offers to a physician a predefined set of the best found classifiers, to allow him an effective work in his every-day practice. The method is semi-automatic - it still leaves to the operator, beside the classifier choice, a possibility to manually (with the mouse) adjust the main parameter (s) which visually, on the fly, grows/shrinks the extracted fibrous region.