A multi-view approach to Consensus Clustering in multi-modal MRI

It has been shown that the combination of multi-modal MRIimages can improve the discrimination of diseased tissue. Thefusion of dissimilar imaging data for classification and segmentationpurposes however, is not a trivial task, as thereis an inherent difference in information domains, dimensionalityand scales. This work proposes a multi-view consensusclustering methodology for the integration of multi-modalMR images into a unified segmentation of tumoral lesions forheterogeneity assessment. Using a variety of metrics and distancefunctions this multi-view imaging approach calculatesmultiple vectorial dissimilarity-spaces for each MRI modalityand makes use of cluster ensembles to combine a set of unsupervisedbase segmentations into an unified partition of thevoxel-based data. The methodology is demonstrated in applicationto DCE-MRI and DTI-MR, for which a manifold learningstep is implemented in order to account for the geometricconstrains of the high dimensional diffusion information.