Please confirm you are human (Sign Up for free to never see this)
← Back to Search
Volumetric Texture Description And Discriminant Feature Selection For MRI
Published 2003 · Computer Science
This paper considers the problem of texture description and feature selection for the classification of tissues in 3D Magnetic Resonance data. Joint statistical measures like grey-level co-occurrence matrices (GLCM) are commonly used for analysis texture in medical imaging because they are simple to implement but are prohibitively expensive to compute when extended to 3D. Furthermore, the issue of feature selection which recognises the fact that some features will be either redundant or irrelevant is seldom addressed by workers in texture classification. In this work, we develop a texture classification strategy by a sub-band filtering technique similar to a Gabor decomposition that is readily and cheaply extended to 3D. We further propose a generalised sequential feature selection method based on a measure of feature relevance that reduces the number of features required for classification by selecting a set of discriminant features conditioned on a set training texture samples. We describe and illustrate the methodology by quantitatively analysing a variety of images: synthetic phantom data, natural textures, and MRI of human knees.