Registration of brain images by a multi-resolution sequential method

Abstract

In many circumstances the clinical interpretation of an imaging from a single image modality is inadequate. A number of registration techniques have been introduced in the literature in order to correlate the clinical information obtained from two different imaging modalities. All methods based on some distance measure suffer from the presence of multiple local minima when minimization algorithms are used to reduce the distance between two edges, or surfaces. A multi-resolution technique has been developed in conjunction with a sequentially improved distance function in order to register sets of MRI, PET, and SPECT images. A global search is initially performed on coarse resolution 3D surface images of each modality where a variable threshold is used to select any likely match location for finer resolution levels. An adaptive termination of the computation of the distance function is possible due to the sequential nature of its evaluation. The superimposed images of MRI and HMPAO images, displayed as slices and in 3-D, were clinically helpful.