The Study Of Computer Aided Diagnosis Of B-ultrasonic Medical Images Of Liver Cancer

The computer-aided diagnosis of medical image is a branch of image processing.More and more scholars and medical experts work at this domain and make great progress.In this paper we studied in the field of B-ultrasonic medical images of liver cancer.We proposed a way which combines with the methods,the Kernel Principle Components Analysis(KPCA)and Rough Sets, to analyze the B-ultrasonic medical images of liver,and we can recognize the B-ultrasonic medical images of liver cancer correctly.The follows is the work of our study:1.Feature images extraction of the B-ultrasonic images of liver cancer.The analysis and measurement were often directly by extracting one or a few characteristics of the original medical images in the most existing methods.In this paper,based on the natural characteristics of the B-ultrasonic images of liver cancer,the feature images were extracted first and then the feature matrix which made of the mean and variance of the feature images,instead of the original image characteristics were analyzed and processed.2.Feature fusion and selection using the method of KPCA.In this paper,the feature matrix was fused by the method of KPCA(Kernel Principle Components Analysis).First select the kernel function and its parameters.Then determine the kernel principle components.As the kernel principle components have a one-to-one relationship with the characteristics of the original images,it was determined that the process of selecting the kernel principle components is the process of selecting the characteristics of the images.Through simulation analysis,the key characteristics which affected the original images most is gray degree and the characteristics which reflect the spatial distribution of gray degree,which provide an important basis for identification.3.Image identification using Rough Sets.Rough Sets is a theory to deal with imprecise, incomplete and uncertain data.One excellent advantage of Rough Sets is that it can analyze the imprecise and imperfect information and classify it accurately when there is no transcendent knowledge.And it can analyze data and reasoning which has implied knowledge and potential rule.Based on the result of the feature fusion and selection,the gray degree and the texture statistical features were determined the attribute sets of Rough Sets which including the skewness,the kurtosis,the average value,the standard deviation,smoothness,the uniformity and the entropy.Then the rules of the classification were made and carried on the recognition to the test samples and made good progress.The accuracy of the recognition to the B-ultrasonic images of liver cancer and the natural liver was surpassed 90%,which achieved the same level or higher rate in the similar research. Compared with the method based on the wavelet textural feature extraction and the artificial neural networks recognition,the method of our paper not only did not need the previous learning process,not affected by the change of the inputs,but also had good stability and was easy to realize.