Towards Application-specific Multi-modal Similarity Measures: a Regression Approach.

In multi-modal registration, similarity measures based on intensity statistics are the current standard for aligning medical images acquired with different imaging systems. In fact, the statistical relationship relating the intensities of two multi-modal images is constrained by the application, defined in terms of anatomy and imaging modalities. In this paper, we present the benefits of exploiting application-specific prior information contained in one single pair of registered images. By varying the relative transformation parameters of registered images around the ground truth position, we explore the manifold described by their joint intensity distributions. An adapted measure is fitted using support vector regression on the training set formed by points on the manifold and their respective geometric errors. Experiments are conducted on two different pairs of modalities, MR-T1/MR-TOF and MR-T1/SPECT. We compare the results with those obtained using mutual information and Kullback-Leibler distance. Experimental results show that the proposed method presents a promising alternative for multi-modal registration.     «

In multi-modal registration, similarity measures based on intensity statistics are the current standard for aligning medical images acquired with different imaging systems. In fact, the statistical relationship relating the intensities of two multi-modal images is constrained by the application, defined in terms of anatomy and imaging modalities. In this paper, we present the benefits of exploiting application-specific prior information contained in one single pair of registered images. By...     »