A continuous and probabilistic framework for medical image representation and categorization

Abstract

This work focuses on a general framework for image representation and image matching that may be appropriate for medical image archives. The proposed methodology is comprised of a continuous and probabilistic image representation scheme using Gaussian mixture modeling (GMM) along with information-theoretic image matching measures (KL). The GMM-KL framework is used for matching and categorizing x-ray images by body regions and orientation. A 4-dimensional feature space is used to represent the x-ray image input, including intensity, texture (contrast) and spatial information (x,y). Unsupervised clustering via the GMM is used to extract coherent regions in feature space, and corresponding coherent segments (“blobs”) in the image content. The blobs are used in the matching process. A dominant characteristic of the radiological images is their poor contrast and large intensity variations. This presents a challenge to matching between the images and is handled via a post-processing stage that provides an invariant blob-signature per image input. In a leave-one-out procedure, each image out of 851 is used once as a test-image, and is categorized by the remaining (labeled) images. The GMM-KL classifier was tested using 851 radiological images with error-rate of 1%. The classification results compare favorably with reported global representation schemes, such as histograms.