Evaluation of statistical features for medical image retrieval

Abstract

In this paper we present a complete system allowing the classification of medical images in order to detect possible diseases present in them. The proposed method is developed in two distinct stages: calculation of descriptors and their classification. In the first stage we compute a vector of thirty-three statistical features: seven are related to statistics of the first level order, fifteen to that of second level where thirteen are calculated by means of co-occurrence matrices and two with absolute gradient; finally the last eleven are calculated using run-length matrices. In the second phase, using the descriptors already calculated, there is the actual image classification. Naive Bayes, RBF, Support VectorMachine, K-Nearest Neighbor, Random Forest and Random Tree classifiers are used. The results obtained applying the proposed system both on textured and on medical images show a very high accuracy. © 2013 Springer-Verlag.