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

Content-based image retrieval (CBIR) systems are used to retrieve relevant images from large-scale databases. In this paper, a framework for the image retrieval of a large-scale database of medical X-ray images is presented. This framework is designed based on query image classification into several prespecified homogeneous classes. Using a merging scheme and an iterative classification, the homogeneous classes are formed from overlapping classes in the database. For this purpose, the shape and texture features, selected using the forward selection algorithm, are optimized by a novel genetic algorithm-based feature reduction and optimization algorithm in the feature space. In this algorithm, using a new fitness function, we try to locate similar images in the database together in the feature space. Using the merging-based classification, the m-nearest classes to the query image are selected as a filtered search space. To increase the retrieval efficiency, we integrate a novel dependency probability-based relevance feedback (RF) approach with the proposed CBIR framework. The proposed RF uses a synthetic distance measure based on the weighted Euclidean distance measure and Gaussian mixture model-based dependency probability similarity measure of the database images to the Gaussian mixture distribution function of the positive images. The experimental results are reported based on a database consisting of 10,000 medical X-ray images of 57 classes (ImageCLEF 2005 database). The provided results show the effectiveness of the proposed framework compared to the approaches presented in the literature.