Multispectral Tissue Identification In MR Images
Michael Merickel, John W. Snell, Theodore R. Jackson, Danny M. Skyba, W.K. Katz · 2005
Magnetic Resonance Imaging (MRI) is rapidly becoming accepted as a valuable diagnostic tool in modem day radiology. One of the most valuable aspects of MRI is that it provides multidimensional (i.e., multispectral) imagery which emphasizes different soft tissue characteristics. This paper describes our work regarding the development and exploration of techniques for the identification and characterization of tissues in such multispectral images. We have employed multiple approaches to this multispectral pattern problem which include the development of classifiers which utilize statistical and morphological information as well as artificial neural networks. The application of these because it permits the disease process to be followed over time in response to therapy, and provides quantitative information permitting the disease to be staged which is essential for surgical planning. In the case of MRI, the ability of the operator to control extrinsic pulse sequence parameters means that overlapping (multispectral) image planes can be obtained. Such MR images creates with different pulse sequences emphasize different tissue characteristics which provides a wealth of multispectral information which can be utilized for tissue identification and characterization. We have been interested in developing image processing and pattern recognition techniques for combining and quantifying information regarding different tissue types from MR multispectral imagery.