Feature Extraction from NMR Images Using Factor Analysis
Tissue characterization based on NMR parameters is the ultimate goal of NMR imaging. In practice, however, quantitative estimation of specific relaxation components in complex biological tissues is difficult. Noise, limited sampling, and heterogeneity of voxel contents are the main reasons why neither multiexponential nor simplified descriptions of relaxation behavior are entirely relevant. Alternatively, the changes in tissue contrast in a series of NMR images can be analyzed by multivariate methods. Satellite image processing, multispectral analysis, cluster analysis, analysis of principal components, fuzzy set theory, and different types of classification algorithm have been used to identify specific tissue NMR signatures and to improve tissue classification (Vannier et al. 1985; Jungke et al. 1988; Ortendahl and Carlson 1988; Pearlman et al. 1988). In contrast to these methods, factor analysis is not directed at classification but rather at a simplified description of tissues by means of a few parameters which may or may not have a physical meaning. The aim of this contribution is to discuss the relevance of a factor analysis model in NMR imaging and the possibility of physical interpretation of the model parameters.
Unable to display preview. Download preview PDF.
- Jungke M, von Seelen W, Bielke G, Meindl S, Grigat M, Pfannenstiel P (1988) A system for the diagnostic use of tissue characterizing parameters in NMR-tomography. In: de Graaf CN, Viergever MA (eds) Information processing in medical imaging. Plenum, New York, pp 471–481Google Scholar
- Ortendahl DA, Carlson JW (1988) Segmentation of magnetic resonance images using fuzzy clustering. In: de Graaf CN, Viergever MA (eds) Information processing in medical imaging. Plenum, New York, pp 91–106Google Scholar