Research On Medical Image Segmentation With Applications In IGRT System

Nowadays, the computer imaging technology is usually used to assist clinical diagnosis ofthe target organ and tissue area. Besides, extraction the data information of target organ andtissue accurately is helpful for quantitative analysis to assist therapy. The crisis organs near thetumor region need to be delineated accurately to guarantee the radiation level on the crisis organsin a reasonable scope during the implementation of radiation therapy plan. Currently, the regionof interesting is usually delineated by radiation therapist and medical physicist in clinic, or theinitial region is segmented by the automatic computer algorithm firstly and then the obtainedinitial region is modified by radiation therapist and medical physicist to get accurate targetregion.The advantage of the manual segmentation is to make full use of the experience andunderstanding about the anatomical structure of the radiation therapist and medical physicist.They are able to translate the experience and understanding about the anatomical structure intopriori knowledge of the crisis organs to achieve segmentation effectively and make the manualsegmentation result accurate and robust. The disadvantage of manual segmentation is highsubjectivity and very time-consuming. In particular, with the development of clinically medicalimaging technology and the demand for better therapy effect, more and more kind of medicalimages are obtained during the clinical treatment, besides the spatial resolution of the medicalimage is bigger and bigger, therefore the very large database is formed. It is time-consuming tosegment the data by traditionally manual segmentation method,meanwhile, the segmentationaccuracy may influenced by the subjective factors.An automatic liver segmentation algorithm based on the characteristic of the medicalimage is proposed in this paper to over come the issues in liver segmentation like liver and itsneighboring tissues and organs have similar intensity distribution and the liver shapes varyhighly across different patients. In this paper, we propose an automatic method to initialize thelevel set based method, which improve the reliability and efficiency of the segmentation method.Then a correlation map is constructed based on the texture analysis within the liver regionobtained by the level set based method. The correlation map calculates the local correlation degree of neighboring pixels, which combined with the level set based method to refine the liverregion.The medical images in radiotherapy planning system have some characteristics, first of all,there are large amounts of historical patient cases in the system. Secondly, all the cases inradiotherapy planning system are made radiotherapy plans, so all the tumor target and relatedcrisis organs have been delineated accurately. Therefore, the segmentation results of the previouscases can be regarded as prior knowledge, which is combined with the feature of current medicalimage to get the final result. The segmentation method based on map is robust when it is used tosegment target region, because the map take use of the huge amount of prior information. Theaccuracy of segmentation by the automatic segmentation method based on map is affected by theregistration accuracy from our study. However, the reference image and target image need to beregistrated in all segmentation method based on map. Therefore, the accuracy and robustness ofthe segmentation method based on map largely depends on which of the registration algorithm.So a deformable registration frame, which detect and mach feature points automatically, ispresented to construct the map. An probabilistic map constructed is used to segment target regionin radiotherapy planning system to assist radiation therapist and medical physicist make radiationtherapy plan.