Based On LSSVM Application Of Tumor Image Classification

Currently in the medical field,we pay more and more attention on the tumor image classification .However, due to the characteristics of the tumor image itself: between the cells and glands, between cells and cells, between the gland and the gland. its adhesion is very serious, leading to the tumor image classification results are not very good. In this paper, through the study of many mature on the basis of classification and recognition method ,we present an improved algorithm which is suitable for image classification and identification of tumor. The research is show as follows:First,for the feature extraction we mainly obtain the image features from the following two ways. on the one hand we total the number of glands in each image as the tumor characteristics: the perimeter of the gland, the average perimeter, area and average size of the gland. On the other hand we consider the number of pixels in each image as the feature (the size of each picture is 266 * 200, or 53200-dimensional.) In the feature reduction process,we overcome the traditional linear PCA and the LDA algorithm characteristics by introducing the kernel function to process the original samples(especially the non-linear characteristics of the sample), and then through the existing PCA and linear discriminant analysis was used to generate lower-dimensional model. Experiments show that the kernel function by introducing a reasonable dimension reduction algorithm can be effectively reduced dimensional model.Second, for the image classification problem we mainly use the improved support vector machine algorithm that is least squares support vector machine (LSSVM) for classification. Compared to the traditional support vector machine, we could avoid arising quadratic programming problems according to the use of LSSVM method.By introducing the kernel function, we could translate the inequality problem of the original quadratic programming into the equation problem, it will not only reduce the amount of its operations, but also improve the operation speed. The results show that: When using the aforementioned reduced-dimensional model to classify the test samples, we can see that using this improved LSSVM method can obtain better tumor classification of the image.In order to confirm the method has a better classification of test samples, we combine the support vector machine (LSSVM) Classification to the related MATLAB functions, and apply it to the tumor image system. Experiments show that using this CAD system can be more clear, more direct .It can represent that the method is so effective that the pathologist can use the auxiliary tool for better research.