3D multi-modality medical image registration using feature space clustering

Abstract

Abstract. In this paper, 3D voxel-similarity-based VB registration algorithms that optimize a feature-space clustering measure are proposed to combine the segmentation and registration process. We present a unifying de nition and a classi cation scheme for existing VB matching criteria and propose a new matching criterion: the entropy of the grey-level scatter-plot. This criterion requires no segmentation or feature extraction and no a priori knowledge of photometric model parameters. The effects of practical implementation issues concerning grey-level resampling, scatter-plot binning, parzen-windowing and resampling frequencies are discussed in detail and evaluated using real world data CT and MRI .

Optimization of mutual information for multiresolution image registration

“…Thus, their scheme does not benefit from the simplified expression of derivatives that we presented at Section IV, which is particularly relevant when computing the second-order derivatives needed for the Hessian. While Viola and Wells produce an estimate of the joint histogram that is entirely continuous, thanks to Parzen windows and thanks to the direct use of unquantized grey values, Collignon et al [13], [15], [28] represent the joint histogram in an essentially discrete fashion: they use binning with regularly spaced bins. Our work is a compromise between these two extreme views, because our representation of the histogram is continuous, like in Viola's approach, and at the same time it is described by a set of discrete and regularly-spaced intensity values, like in Collignon's approach.…”

Clinical validation of the normalized mutual information method for registration of CT and MR images in radiotherapy of brain tumors

“…An alternative approach is to derive measures from (the spatial distribution of) the respective voxel intensities of each data set, which are then employed as parameters for registration methods. (7)(8)(9)(10)(11)(12)(13)(14) Studholme et al (16) compared four such similarity measures and concluded that measures of soft tissue correlation (11) and mutual information (13,14) provided robust solutions for the registration of 3D CT and MR data sets. Freire et al (22) also demonstrated the robustness of the Geman-McClure estimator in a comprehensive study of several intensity-based measures.…”

Combining landmark and intensity driven registrations

“…Intensity based approaches aim to match images by minimizing an appropriate distance measure, like, e.g., the L 2 -norm of the difference image or the mutual information of the two images; see, e.g., Brown [4], Collignon et al [6], Roche [15], or Viola [19]. Based on these distance measure, a variety of registration techniques has been developed; see, e.g., D'Agostino et al [7], Hermosillo [10], or Modersitzki [14].…”

“…For this measure to be successful, one has to assume that the intensities of the two given images are comparable. Other distance measures, capable of dealing with multimodal images, like, e.g., mutual information (cf., e.g., Collignon et al [6] or Viola [19]), are also under consideration; see, e.g., D'Agostino et al [7], Hermosillo [10], or Roche [15].…”