A Study On Medical Image Registration Based On Shape Information

Medical image registration is a kind of image processing technology that develops rapidly in recent years. Registration of medical images from different imaging devices or different time could be used for medical diagnosis, surgery planning, tracking of pathological changes, treatment assessment, and so on.In general, feature-based image registration has low computational complexity and fast speed, but low accuracy; whereas, intensity-based image registration is more accurate and robust, but has high computational complexity and slow speed. In order to take advantage of both of them, we proposed a new method which combined the maximum mutual information optimization with shape matching: first, we extracted the target shapes from reference image and floating image and performed matching; then, we carried on the mutual information optimization with the initialization calculated from the matched landmarks. Experimental results indicated that the new algorithm had low computational complexity, fast speed, high accuracy, and could avoid the convergence at the local extremum when performed parameters optimization, so it improved the speed and accuracy of the registration effectively.Image Registration needs large amount of calculation, and the speed becomes one of key factors in practical application. In order to accelerate the registration process, we presented an optimization method based Intel SIMD technology, which exploited the internal parallelism of the registration process, and adopted a SIMD data organization to implement the parallel computing. The experiment result demonstrated the proposed method accelerated the speed compared with the conventional implementation in some parts of registration, giving a good prospect in practical application.