Multi-modal Medical Image Registration Based On Structure And Regional Mutual Information

With the rapid development of science and technology, various sensors are applied to many areas, it makes the medical imaging technology develop steady. Now, many technologies are used to medical imaging such as CT, MR, X-ray, ultrasound and so on, the images obtained from each imaging technology have their own characteristics. And with the need of the development of the medical technology, the data is simple that provided by the single-modal medical image, it often cannot reflect the characteristics of the lesions accurately, cannot provide enough data to help doctors make better diagnosis. Therefore, fusing the multi-modal image features to help doctors diagnose illness becomes an important problem. But the premise of the multi-modal image fusion is multi-modal image registration, so getting the accurate registration results are particularly important.The image registration methods can be divided into three types approximately: one is based on the gray information, the second is based on features, and the third is based on the transform domain. The method that is based on gray information do not need to process image beforehand, the implementation process is simple. The method which is based on feature should preprocess the image before registration, extract the image feature, and according to the image feature to register. Based on the transform domain image registration is done in frequency domain. But because of the emphasis of the multi-modal image is different, it is easy to cause the false registration. The structure information obtained from the same part of the multi-modal image is consistent, extracting the accurate image feature is important to multi-modal image registration, so, the focus of the paper is combining the structure information to complete the registration. In this paper, the research work is divided into follow parts:First, extract the phase congruency image, phase congruency is a method of image feature detection, do the Fourier transformation to the image first, then regard the point that the phase of its Fourier component is the most consistent as the image feature, this method can obtain a complete image feature.Second, obtain the gradient amplitude image, and fuse the phase congruency image and the gradient amplitude image reasonably, thus get the relatively complete structure information image.Finally, use the regional mutual information to register, and genetic algorithm to achieve optimizing, it is not only used the structure information of the image, but also combine the gray information of the image, it makes the registration more accurate, more effective, and the stability is better.Verified by the experiment, extracting image feature base on the phase congruency is not affected by the noise, light and contrast. Combining the phase congruency image with gradient amplitude image can get the clear structure information image, the registration that based on the structure information and regional mutual information is not only high precision, high efficiency, and good robustness.