A Texture Dictionary for Human Organs Tissues’ Classification
Daniela Stan Raicu, Jacob David Furst, David S. Channin · 2004
The research presented in this paper is expected to aid the process of medical decision making by providing tools for automatic extraction of the most discriminative features of regions of interest in medical images produced by the Computerized Tomography (CT) modality. The regions of interest studied in this paper are the liver, heart, backbone, kidneys, and the spleen. To characterize the regions of interest, we use texture information as well as textual information. We capture the texture information of the regions of interest using the Haralick texture descriptors and the run-length encoding descriptors. The texture information is given by the keywords annotating the organs. We apply Latent Semantic Indexing (LSI) to find the relationships between the texture features and the header information; the motivation behind using LSI is the cross modality ability of the technique that allows the combination of different kinds of data in discovering the relationships. These relationships are stored in a Texture Dictionary that can be later used to automatically annotate new CT images with the appropriate organ names.