Partial clustering for tissue segmentation in MRI

Abstract

Magnetic resonance imaging (MRI) is a imaging and diagnostic tool widely used, with excellent spatial resolution, and efficient in distinguishing between soft tissues. Here, we present a method for semi-automatic identification of brain tissues in MRI, based on a combination of machine learning approaches. Our approach uses self-organising maps (SOMs) for voxel labelling, which are used to seed the discriminative clustering (DC) classification algorithm. This method reduces the intensive need for a specialist, and allows for a rather systematic follow-up of the evolution of brain lesions, or their treatment. © 2009 Springer Berlin Heidelberg.