Evaluation of Medical Image Registration by Using High-Accuracy Image Matching Techniques
An effective method for quantitatively evaluating deformable image registration without any manual assessment is developed in the work of this chapter. Fiducial landmarks for spatial evaluation are firstly detected using a feature-point detector in a fixe
- PDF / 488,623 Bytes
- 20 Pages / 439.37 x 666.142 pts Page_size
- 33 Downloads / 196 Views
Abstract An effective method for quantitatively evaluating deformable image registration without any manual assessment is developed in the work of this chapter. Fiducial landmarks for spatial evaluation are firstly detected using a feature-point detector in a fixed image, and corresponding points in the registered moving images are localized with a high-resolution image matching algorithm. Distance between the reference points and the correspondences can be used to estimate image registration errors. With the developed method, users can evaluate different registration algorithms using their own image data automatically.
Introduction Image registration aims to find a spatial transformation that maps points from one image, a moving image, to corresponding points in another image, a fixed image. Medical image registration is fundamentally used in many applications, such as diagnosis, planning treatment, guiding treatment, and monitoring disease progression. Thus, it is necessary to validate whether a rigid/deformable registration algorithm satisfies the needs of an image processing application with high accuracy, robustness, and other performance criteria. The most straightforward method for estimating image registration error is to compare a given registration transformation with a “gold standard” transformation [1], whose accuracy is high. However, the lack of a gold standard prevents any automatic assessment of registration accuracy. An attempt that stands out in this regard is the “Retrospective Image Registration and Evaluation (RIRE) project” [2] for evaluation of brain-image rigid registration. The RIRE project used boneimplanted fiducial markers to obtain a marker-based rigid transformation as the Z. Li (*) • T. Kurihara Intelligent Media Systems Research Department, Central Research Laboratory, Hitachi, Ltd., 1-280 Higashi-Koigakubo, Kokubunji-shi, Tokyo 185-8601, Japan e-mail: [email protected] A.S. El-Baz et al. (eds.), Abdomen and Thoracic Imaging: An Engineering & Clinical Perspective, DOI 10.1007/978-1-4614-8498-1_19, © Springer Science+Business Media New York 2014
489
490
Z. Li and T. Kurihara
gold standard transformation. Registration error was measured by calculating the error relative to the gold standard over a set of specified regions. For deformable image registration, synthetic images and phantoms were utilized as reference standards to provide qualitative evaluation of registration performance [3–6]. However, such standards lack sufficient realism, compared with that derived from actual patient image data. “Non-rigid Image Registration Evaluation Project (NIREP)” in [7] and recent work in [8, 9] provided intensity-based metrics for evaluating the registration performance of brain images, using manually segmented anatomical regions from actual clinical data. These projects required manual annotation and segmentation to create evaluation databases, and the evaluation data only included brain images. On the other hand, a number of studies utilized expert-determined landmark features t
Data Loading...
