Interpolative decisions in the fuzzy signature based image classification for liver CT

In computer aided diagnostics image processing and classification plays an essential role. Image processing experts have been developing solutions for different types of problems, that can be related to image processing, however, due to the sensitivity of the data and the high cost of medical experts, these experimental methods usually have very limited use in real medical practice. The databases that are available are very limited, thus the elsewhere usual and extremely effective convolutional neural network or other automated learning methods are not so easy to adjust for medical image processing. To overcome this difficulty, this paper presents an expert knowledge based method which describes the decision structures by fuzzy signatures. Values of various properties of Computer Tomography images as e.g. density or homogeneity are being considered in these signatures that are different in all case of liver diseases. Because of the low number of samples available, fuzzy sets that describes the leafs of the signatures leads to sparse systems, hence interpolation is needed. However further investigations of other interpolation methods are planned, Stabilized Koczy-Hirota interpolation seems to be appropriate.