Eigenspace normalization of multi-spectral magnetic resonance images

The most frequently used imaging procedure for the diagnosis of diverse neurological illnesses is magnetic resonance. Improved benefits have been obtained from multi-spectral information under this imaging modality. Both supervised and unsupervised segmentation procedures have been reported for this kind of images, many of which require a preprocessing stage for contrast adjustment within the intensity space. This paper presents a fully automatic procedure for intensity space normalization of multi-spectral magnetic resonance image stacks based on the combination of the Karhunen-Loeve transformation and a maximum likelihood adjustment of the basis vectors' orientation as related to an underlying mixture model available for supervised segmentation. Results on three different test cases are presented, together with the Tanimoto indexes computed for segmented sets of images from each case. The average indexes obtained qualify the normalization procedure as an adequate alternative for multi-spectral magnetic resonance image preprocessing.