Research On Medical Image Segmentation In TPS System

As a non-interposition therapy mean, radiotherapy technology is a very important way for malignant tumor treatment. Before we actualize the radiation to the patient, we must plan the therapy and simulate the process of the treatment. With the progress of the visualization technology, medical visualization technology is playing an important part in the therapy planning of the treatment.TPS (Treatment Planning System) makes therapy planning quantificationally and optimizes it to make the planning accurately. Image processing technology is an important part of TPS system and it is one pivotal technology of it. The precise segmentation of medical image with tumor makes the 3D reconstruction correctly and makes the therapy planning accurately. It plays an important part in TPS system.In order to carry out a tumour therapy planning with high-precision, high-dosage, good curative effect and low damnification, we need to draw the outline of the tumour and natural tissue around it correctly. It brings a high requirement of image segmentation. There are different focuses in kind and shape in clinic, and many different segmentation methods, so we need to choose appropriate segmentation scheme according to the characteristic of the focus image and the property of the segmentation method. Most of the medical image segmentation methods are applied to the recognition and classification of the whole image, we can't get high precision of the focus and important organ segmentation with these methods.Based on the theories of morphology, mutual information, medical image processing and the demand of the clinic, this thesis makes a deeper research on main technologies of medical image segmentation in TPS system. The main contents of this thesis are taken as follows:Firstly, morphology theory and its application in image filter is studied in this thesis. We research the function of different morphological operator in filter application. Combination of the open and close operation can reduce noise and smooth the image, but the open-close filter can't get fine effect if the image with great noise. In order to reduce noise highly, an improved method of open-close morphology filter is provided, and then it is combined with improved median filter to reduce noise. Several medical images with different noise intensity are simulated in MATLAB using the improved filter method. The results of experiment show that the effect of filter is satisfying, the noise is obviously weakened.Secondly, this thesis gives a new method, HIST-MI, of medical image refined segmentation on the basis of the theory of mutual information and the application of it in medical image processing. This method is used to segment object which needed to be segmented accurately in TPS system. We use mutual information as the threshold estimation and simulate the morphological watershed method, to find a threshold vector T. First, histogram of the medical image was obtained. Second, the histogram was considered as a mountain containing ridge and valley, and it was under the water surface. Then, made water down, the point that showed up first was regarded as the first object, the second point was judged to belong to the same object if it was adjacent to the first point, otherwise it belonged to another object, and suchlike. In order to reduce the calculation, HIST-MI method was combined with Otsu method to segment image. Several medical images are simulated in MATLAB using the HIST-MI method. Good results were obtained and this method was not sensitive to the noise. It could be used to segment the important organ and the pathological tissue in the TPS system.Finally, we make a conclusion and propose the future research directions in this field.