A framework for medical image retrieval using merging-based classification with dependency probability-based relevance feedback

摘要

Content-based image retrieval (CBIR) systems are used toretrieve relevant images from large-scale databases. In this paper, aframework for the image retrieval of a large-scale database of medicalX-ray images is presented. This framework is designed based on queryimage classification into several prespecified homogeneous classes.Using a merging scheme and an iterative classification, thehomogeneous classes are formed from overlapping classes in thedatabase. For this purpose, the shape and texture features, selectedusing the forward selection algorithm, are optimized by a novelgenetic algorithm-based feature reduction and optimization algorithmin the feature space. In this algorithm, using a new fitnessfunction, we try to locate similar images in the database together inthe feature space. Using the merging-based classification, them-nearest classes to the query image are selected as a filtered searchspace. To increase the retrieval efficiency, we integrate a noveldependency probability-based relevance feedback (RF) approach with theproposed CBIR framework. The proposed RF uses a synthetic distancemeasure based on the weighted Euclidean distance measure and Gaussianmixture model-based dependency probability similarity measure of thedatabase images to the Gaussian mixture distribution function of thepositive images. The experimental results are reported based on adatabase consisting of 10,000 medical X-ray images of 57 classes(ImageCLEF 2005 database). The provided results show theeffectiveness of the proposed framework compared to the approachespresented in the literature.