The Study And Application Of New Segmentation Method On Medical Ultrasonic Image

Medical ultrasound imaging, computerized tomography, magnetic resonance imaging and nuclear imaging are the four important medical imaging techniques in modern society. Among them, medical ultrasound imaging is widely used in medical diagnosis and treatment due to its unique characteristic of real-time, noninvasiveness, cheapness, easy repetition, and high sensitivity. It has high potential in quantitative analysis, real-time monitoring and treatment scheduling, etc. When ultrasonic images are utilized to do medical diagnosis, it is necessary to extracting the region of interest for further analysis' convenience. This process is so-called image segmentation. Medical ultrasonic image segmentation has been a classical tough problem since many years before. The main reason is that the resolution and contrast degree of the ultrasonic image itself is very low as well as the influence of intrinsic speckle noises. Based on the fact above, this thesis is developed to do some research on ultrasonic image segmentation to improve current algorithms or propose new methods.This thesis first gives an overview of the ultrasonic image segmentation methods. Then several commonly used segmentation algorithms are analyzed with reference to the experimental results on the ultrasonic images. As to the complicated ultrasonic image, we may foresee the dissatisfied results with every one of these simple theory and methods. This also can be confirmed from the results of experiment. So making effectively use of their respective advantages, and integrating with other image analysis tools such as genetic algorithm, wavelet multi-scale analysis, math-morphology theories and so on is a new research direction. Still there is more latent potential to tap for these methods on ultrasonic image segmentation.On the basis of the work of the former researchers, this thesis subsequently develops the work from the following two aspects. Firstly, integrating with theories such as wavelet multi-scale decomposition, the author proposes a new segmentation method based on scale co-occurrence matrix. The advantages are affirmed in the thesis by comparing with gray-level co-occurrence matrix, which proposed by Wang. Secondly, integrating EMD theories, LAWs texture energy measure, fuzzy clustering and math-morphology theories this thesis proposes another image segmentation new method, and successfully extracted the region of illness. This is the first EMD application on ultrasonic image segmentation.