Research And Application On The Technique Of Computer Pattern Classification Based On Medical Image

This research work is a part of the Research of Data Mining on Medical Image supported by the National Natural Science Foundation of China. Some key techniques and main algorithms of computer pattern classification for the mammograms database is proposed and a computer-aided medical diagnosing system on breastcancer is also developed. The main work is as follows:(1) Image PreprocessingSince images are polluted by noise during transferring and transforming,preprocessing of the images is necessary to improve the quality of the images and to make the feature extraction more reliable. A simple and effective median smooth filter with boundary holding is applied to eliminate noise in the digital mammograms. At the same time, histogram equalization is used to implement image enhance andimprove qualitative.(2) Gray-Level Co-Occurrence Matrices and Texture FeaturesTexture features of image describe the local patterns and the arranging rulesappearing through and through in the image. These texture features macroscopically reflect some rules of gray changing. In the present study, four GCMs are constructed in four directions and thirteen texture features independent of directions areextracted. (3 ) Detection of the Breast Tumor boundary and Features ExtractionA region growing method to extract accurate boundary of the breast tumor regionwas investigated, and the corresponding compactness, moment and Fourier descriptors are extracted. Experimental results show that the above shape factors can describe the shape of breast tumor perfectly and are very effective in distinguishingthe benign from the malignant tumor. (4) Classifier DesignA proximal support vector machine (PSVM) classifier is proposed in terms of deeply study of support vector machine (SVM) ,which not only runs faster than standard support vector machine classifiers but also is easy to implement withsatisfactory result for lower hardware. As an active branch of machine learning, the artificial neural networks(ANN) is applied to many fields, such as intelligent control, pattern recognition and signal process. The principle and standard algorithm of the neural networks is deeply analyzed with its advantages and disadvantages in this thesis. Then the improved neural networks is proposed and experiment results show that it is a very perfect classifier.