Probabilistic intensity mapping in MRI image registration

In this work, we present a method which is able to relate different MR sensors with respect to intensity distortions in the output images. For the important problem of image registration, the method makes possible a principled approach to likelihood modeling or the construction of similarity metrics. Likelihood models can be used as prior knowledge of the relationship between intensities in both images, providing a fundamental information resource for image registration. A poor model of the intensity mapping for the image pair to be matched may lead to false matches, regardless of the prior morphological constraints assumed and will bias all subsequent analyses. A formal analysis of robustness under different kinds of noise is also provided and the findings compared to other relevant similarity metrics. Experiments are controlled based on the application of synthetic spatial and intensity deformations that guarantee a fiducial basis for comparison.