Minimum distance processor for biological tissues classification from A-scan ultrasonic signals

A new technique for biological tissue classification is presented. The classification problem was to find the correct tissue type based on the observed data vectors which were assumed to consist of the true underlying backscattered signal and an additive white Gaussian noise due to the measuring system. The power spectrum of the maximum likelihood estimate (MLE) of the backscatter signal was used to classify the different tissue types. Each MLE observation vector was computed from 60 A-scans. Three different biological tissues were used as hypotheses for the classification problem: liver, kidney and pancreas. Using the Bayes criterion and the general Gaussian problem formulation the complexity of the problem was reduced to that of the design of a minimum distance processor by a change of coordinate system. The new coordinate system was computed by the Gram-Schmidt orthogonalization method. Results obtained from the three different tissues (kidney, liver and pancreas) revealed the probability of correct classification at 90%.