Simultaneous registration and tissue classification using clustering algorithms

We describe a novel approach for performing registration and tissue classification of multichannel medical images. Rather than perform a two-step process comprised of a registration step followed by a tissue classification step, the two objectives are accomplished simultaneously using a single algorithm. The new algorithm is based on minimizing a fuzzy C-means clustering energy functional with respect to not only the cluster centers and membership functions, but the transformation parameters as well. The advantage of this simultaneous approach is that both the registration and segmentation now optimize the same cost functional. This approach also allows the registration of more than two images to be easily accommodated. The method is evaluated using both real and simulated magnetic resonance images of the brain.