Unsupervised MRI Tissue Classification by Support Vector Machines

E. Karp and R. Vigário (Finland)


Medical imaging, image processing and signal processing, magnetic resonance imaging, support vector machines, un supervised classification


The objective of this work was to develop better visuali sation tools and techniques for detection and follow up of pathologies in magnetic resonance images. Support vector machines were used, in an unsupervised manner, to seg ment tissues in MR images with different imaging param eters. The segmentation rested on a training set of labelled feature vectors defined using independent component anal ysis. Both simulated and real data was used. Support vector machines proved to be a suitable tool for classification of MR images. The classification error rates for the simulated data indicated that rather good segmentation precision was achieved.

Important Links:

Go Back