With the growing size of medical image libraries, there is a need for efficient tools that can analyze the content of medical images and represent it in a way that can be efficiently searched and compared. The objective of this research is to explore a statistical framework for image representation and image matching in content-based retrieval and analysis of 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, in particular the Kullback Leibler (KL) measure. The GMM-KL framework is used for matching and categorizing x-ray images by body regions. A multi-dimensional feature space is used to represent the x-ray image input, including intensity, texture and spatial information. Unsupervised clustering via the GMM is used to extract coherent regions (“blobs”) in feature space, thus creating Mixtures of Gaussians (MoG) that represent the image content. The MoGs are used in a Nearest Neighbor (NN) retrieval process.
The image matching framework is also extended to include image category modeling and image-to-category matching in order to perform classification tasks and/or speed up the retrieval schemes. Note that only 1% of the number of comparisons in the leave-one-out classification is needed for the category-modeling based classification. Since the models of categories integrate the features from all the images in them, they are not only large but also noisy. Using novel methods based on the Unscented Transform the number of blobs in the categories’ MoGs has been reduced, thus creating compact models for image categories.
The proposed methodology is general and can be extended to additional modalities and labeled categories.
Retrieval example for various organs
For problems or
questions regarding this web page contact