The Research On Key Technologies Of 3D Medial Image Segmentation And Registration For The Demand Of Clinical Applications

In clinical applications, doctors usually like to observe the characteristics of lesion(e.g.tumor) in order to achieve effective diagnosis. During surgery planning, it is necessary t perform the segmentation of vital organs(such as bone, blood vessels, and etc.) that surround the surgical area to avoid damage to these important parts of the body during surgery. In clinical practice, multi-mode image registration can make doctors acquire comprehensive information that combines the superiority of images that come from different devices in order to obtain a higher medical quality. Furthermore, doctors may want to evaluate the result of previous treatment with images from the same patient of different time. In surgical navigation,a successful surgery can be achievedby avoiding important organs with multi-mode image registration.This study focused on the key problems of clinical treatments: medical image segmentation and medical imaging registration. To meet the clinical demand, this paper has made some implementations and improvements of medical image segmentation and registration algorithms, as well as the software. The main frame work is showed as follows:Firstly, the backgrounds of medical image segmentation and several commonly used segmentation algorithm are described in detail, and the advantages and disadvantages of each method are analyzed. The vessel segmentation and bone segmentation are performed by manually selecting seed points. The strategy of multi-seeds selection is proposed to overcome the drawback of the commonly used single-seed segmentation that the segmented region can not be disconnected. The proposed approach of multi-seeds selection can achieve the case of disconnected bones region segmentation.Then, the background of medical image registration and commonly used algorithms are introduced. The registration results of multi-model clinical images(CT and MRI) with rigid transformation and affine transformation are proposed. The checkboard method and volume merge method are also designed to observely evaluate the result of image registration.Finally, the design of software platform for clinical applications is discussed. The software architecture is noverly desgined, and the functions of above proposed segmentation and registration algorithms are all embedded in the designed platform. The whole platform has been tested by clinical images and some results are provided also.