Ching-Fen Jiang and Jia-Yin Li
image registration, mutual information, brain image
Different types of medical images have their own unique characteristics and thus provide different information. Registration techniques integrate the information from various medical imaging from different modalities to align properly in the same coordinates. An efficient and affordable method to register three-dimensional (3D) image data from multiple imaging modalities is in great demand in clinical applications, especially in the evaluation of brain functional changes associated with structural abnormalities. Current auto-registration methods designed for 3D medical images can be divided into two categories: intensity-based and geometry-based methods. The mutual information (MI) algorithm is the most representative method in the first category. We used the previously developed registration method based on contour matching to represent the second category. This method contains two steps: course alignment using principal axes registration (PAR) and fine tuning through maximal cross-section matching. The aim of this study is to compare the performance between these methods. The results from seven sets of head volume acquired by computed tomography (CT), magnetic resonance imaging (MRI) and single photon emission computed tomography (SPECT) show that the proposed contour-matching method is better than the MI algorithm in terms of registration accuracy, especially when the contents between the pair of image volume for registration are inconsistent.
Important Links:
Go Back
