Automatic assessment method for CT-MR image fusion based on a local confidence measure

Validation of current intermodal image registration tech- niques is crucial in clinical practice. This paper presents a novel, au- tomatic CT-MR image registration assessment algorithm that provides the user with an intuitive, colour-coded visualization of the extent and lo- cation of poor registration accuracy of dense structures. This is achieved by using a local confidence measure that compares the mean intensity and variance of small neighbourhoods of MR voxels to an automatically pre-computed expected value. Our goal was to assess the accuracy locally in every region of the image since the level of mis-registration in a given image volume is non-uniform. The algorithm has been tested on CT-MR fused patient datasets of the head, and has produced very encouraging preliminary results, which are discussed in this paper.