Linear spectral unmixing approaches to magnetic resonance image classification.

摘要

Linear Spectral Unmixing (LSU) has been widely used in remote sensing community. By considering Magnetic Resonance (MR) images as multispectral images, LSU can be also applied to MR image analysis. This thesis investigates issues in spectral classification for MR images using LSU approaches. Due to a limited number of MR images acquired by MR sequences, the ability of the LSU cannot be fully explored and utilized. In order to mitigate this dilemma, two approaches are developed to address this issue. One is Band Expansion Process (BEP) to expand an original set of MR images to an augmented set of multispectral images by including additional spectral band images that can be generated from the original MR images using a set of nonlinear functions. In this case, the LSU will have sufficient spectral bands to accommodate signal sources to be unmixed. The other is Kernel-based Principal Components Analysis (KPCA) to expand the original data space to a higher dimensional feature space in which the LSU can perform more effectively. In order to demonstrate the utility of the LSU in MR image analysis, two sets of MR images, synthetic MR images available on website and real MR images, are used for experiments. Experimental results show that the LSU can be a very effective technique in quantifying MR substances to calculate their partial volumes for further MR image analysis.