Classification-Driven Pathological Neuroimage Retrieval Using Statistical Asymmetry Measures

This paper reports our methodology and initial results on volumetric pathological neuroimage retrieval. A set of novel image features are computed to quantify the statistical distributions of approximate bilateral asymmetry of normal and pathological human brains. We apply memory-based learning method to nd the most-discriminative feature subset through image classi cation according to prede ned semantic categories. Finally, this selected feature subset is used as indexing features to retrieve medically similar images under a semantic-based image retrieval framework. Quantitative evaluations are provided. 1 Motivation Medical images form an essential and inseparable component of diagnosis, intervention and patient follow-ups. In this work, we use a patient's image as an index to retrieve medically similar and relevant patient cases from a large multimedia database. Common practice in the image retrieval and pattern recognition community is to map each image into a set of numerical or symbolic attributes called image indexing features. Thus each image corresponds to a point in a multidimensional image feature space. Existing \content-based" image retrieval (CBIR) systems [6, 13] depend on general visual properties such as color and texture to classify diverse, two-dimensional (2D) images. However, these general visual cues often fail to be e ective discriminators for image sets taken within a single domain, where images have subtle, domain-speci c di erences. Furthermore, these global statistical color and texture measures do not necessarily re ect or have proven correspondence to the meaning of an image, i.e. the image semantics, nor are they suitable for handling three-dimensional (3D) volumetric images. Our objective is to go beyond the ill-de ned, subjective visual feature indexing practiced in many current CBIR systems. Our approach is based on statistical learning, and contains the following components: 1. Feature extraction maps each volumetric image into a multi-dimensional image feature space; 2. Feature weighting and image similarity construction imposes relative weights on the image feature space; 3. Adaptive image retrieval chooses image similarity best suited for the user intention. ?? This research is supported in part by an NIST grant #70NANB5H1183 and in part by the NIH/NCI research contract # N01-CO-07119.