Research And Application Of Tissues CT Images Classification Based On KNN

With the rapid development and progress of modern medical imaging technology, especially the inventing of advanced medical equipment such as the CT machine, making a lot of CT medical images of human tissues grow day by day. These medical CT images in diagnosed not only provide beneficial reference value for the doctors at the same time, but also bring them great workload. Because of big challenges and difficulties in medical image processing filed, it emerges many kinds of texture feature analysis technologies and classification techniques on medical image. The texture feature analysis technologies including gray level co-occurrence matrix, wavelet transform, Gabor filter, etc. Classification technologies such as support vector machine(SVM), neural network, Bayesian, decision tree, etc. These methods all have their own application to a certain extent, and there are a lot of researchers making related researches and achievements on CT image classification technology, but they were only on a certain group of human body. The research of multiple organizations CT image is little. Therefore, under the above background, this dissertation is on view of the visceral of human tissue CT images, including the liver and lung lesions and normal image, to doing the a series of researches on texture feature extraction and automatic classification technology.The main contributions of this dissertation are as follows:1. This dissertation proposes PWT-GLCM which is pyramid wavelet transform and gray level co-occurrence matrix. This two methods combined together, first, in frequency domain and time domain, using pyramid wavelet transform method to make three layers decomposition of the preprocessing medical CT image, getting different frequency and different directions of the sub band images; second, in spatial domain, using gray level co-occurrence matrix method to extract and calculate the texture feature parameters of the decomposed sub band images, the results show that the new extraction method PWT-GLCM extracts the texture eigenvalue is better than that using the extraction methods of gray level co-occurrence matrix. 2. This dissertation proposes PCA-KNN classification model which is principal component analysis and K nearest neighbor. As is known to all, the principle of the KNN classification model is to compute the distance between the test temple and the known temple, in the case of multidimensional texture features, KNN classification needs a large amount of calculation time. The results show that the improved KNN classification model in terms of computation time complexity is superior to the traditional KNN classification model. 3. This dissertation designs and implements a human visceral organization CT image classification system prototype. This system focuses on the basic procedures, including image processing, PWT-GLCM texture feature extraction and PCA-KNN classification. The results of the experiment show that PWT-GLCM method extracts the texture feature values is better than that only using GLCM method in texture feature extraction; the computation time of PCA-KNN classification model is lower than the traditional KNN classification model based on the same conditions, which achieves a desired effect.