Gaussian mixture models for semantic ranking in domain specific databases with application in radiology

With recent advances in imaging techniques, huge quantities of domain-specific images, such as medical or geospatial images, are produced and stored daily in computer-based image repositories. Size of databases and limited time at hand makes manual evaluation and annotation by domain experts difficult. In such cases computer based methods can be used to enrich the process of decision making while eliciting previously unknown information. For example, in the medical domain, query by image methods can be used by medical experts for differential diagnosis by displaying previously evaluated cases that contain similar visual patterns. Also, less experienced practitioners can benefit from query-by-semantic methods in training processes, especially for difficult-to-interpret cases with multiple pathologies. In this article we develop a methodology for ranking medical images using customized mixture models. The regions of interest are determined using Dirichlet process to determine natural groupings of images in a content-based feature space. These natural groupings of images are then evaluated for relevance to mixtures of associative semantic mappings. We evaluate and compare the performance of our method on two medical datasets using mean average precision and precision-recall charts.