Research And Implementation On The Registration Of CT And Ultrasound Data

As computer technology develops, clinical medical treatment demands higher standard for medical image process technologies. In this field, doctors often use medical images that are taken from different devices or taken from the same equipment but at different period or angles to make the correct decisions. However, because of the different concepts or the various circumstances, there must be big differences in multi-model images. In order to deal with this problem, the Image Registration technology has been proposed and became focus and difficulty in this field.In order to raise the successful and precise ratio, this thesis studies multi-model images registration problems and proposes a method to optimize the approach based on the Mutual Information. We use an iteration method, which put Non-local Mean Algorithm and Morphology together to extract the contour information and adapt weighted Mutual Information to realize the registration. Finally, the success ratio rises to 96%.In this thesis, the main works and contributions are as described as follows:1. Proposed a method to extract the image outline and make sure the character information reserved and picked up. Through using the Non-local mean algorithm, the overall contents in ultrasound image denoise and enhance the features. At the same time, this paper adopts the morphology gradient algorithm to extract the contour information, and then make preparation to the next step in registration.2. Adopted the Mutual Information Similarity measurement based on the Parzen window; used Powell's algorithm to optimize the searching and employed PV (Partial Volume) algorithm to realize the interpolation. All of above make sure the multi-model image registration success.3. Optimized the traditional Mutual Information registration method, by introducing weighted parameter in the entropy that is the concept in MI (Mutual Information). The value of MI is enhanced and the success ratio and accuracy are improved because the influence by local extrema reduced.For the problems in the image registration based on Mutual Information, this paper studies deeply and proposes a series effective optimized method and realization. And finally, the success ratio and accuracy have better results.