HAMMER: hierarchical attribute matching mechanism for elastic registration

A new approach is presented for elastic registration of medical images, and is applied to magnetic resonance images of the brain. Experimental results demonstrate very high accuracy in superposition of images from different subjects. There are two major novelties in the proposed algorithm. First, it uses an attribute vector, i.e., a set of geometric moment invariants (GMIs) that are defined on each voxel in an image and are calculated from the tissue maps, to reflect the underlying anatomy at different scales. The attribute vector, if rich enough, can distinguish between different parts of an image, which helps establish anatomical correspondences in the deformation procedure; it also helps reduce local minima, by reducing ambiguity in potential matches. This is a fundamental deviation of our method, referred to as the hierarchical attribute matching mechanism for elastic registration (HAMMER), from other volumetric deformation methods, which are typically based on maximizing image similarity. Second, in order to avoid being trapped by local minima, i.e., suboptimal poor matches, HAMMER uses a successive approximation of the energy function being optimized by lower dimensional smooth energy functions, which are constructed to have significantly fewer local minima. This is achieved by hierarchically selecting the driving features that have distinct attribute vectors, thus, drastically reducing ambiguity in finding correspondence. A number of experiments demonstrate that the proposed algorithm results in accurate superposition of image data from individuals with significant anatomical differences.

I. Introduction

Deformable registration of brain images has been an active topic of research for over a decade. Its clinical applications are numerous. In particular, deformable registration is used for spatial normalization of functional images, group analysis, and statistical parametric mapping [1]. It is also used in computational anatomy as a means for measuring structures, by adapting an anatomical template to individual anatomies [2] [3]– [11], [46]. Finally, it is used as a means for image data mining in lesion-deficit studies [12], as well as in stereotaxic neurosurgery for mapping anatomical atlases to patient images [13] [14]– [16]. Therefore, many image analysis methodologies have been developed to tackle this issue, which fall in two general categories. The first family of methods involves feature-based matching, i.e., transformations that are calculated based on a number of anatomical correspondences established manually, semiautomatically, or fully automatically on a number of distinct anatomical features. Such features are distinct landmark points [17], [18], or a combination of curves and surfaces, such as sulci and gyri [6], [19]–[24]. The second family of methods is based on volumetric transformations, which seek to maximize the similarity between an image and a template, and generally assume that the image and the template have been acquired with the same imaging protocol [1], [25]–[30], [47], [48], [51].