| Lung cancer is one of the high incidence malignant tumors, and the inci-dence is increasing year by year. The key to determine whether the patient istreated with surgery is to be clear if there exist metastatic lymph nodes beforeoperation. As the most advanced functional molecular imaging technology, PETdiscovers qualitative lesions earlier than morphological means such as CT, and itis significantly superior to conventional methods in the diagnosis of metastaticlymph nodes. But high sensitivity of imaging agent for PET will bring some falsepositive. Therefore, it’s not scientific to solely rely on SUV value of PET imageto diagnose metastatic lymph nodes.Against PET’s false positive and the limitation of the application of CT inqualitative diagnosis, starting from the texture’s application in the medical imageprocessing, this paper selects the concentrated lymph nodes as the research object,extracts the grayscale space change information of enlarged lymph nodes in CTimages and regional texture change information of enlarged lymph nodes in PETimages, breaks through the restriction of the image morphological information,deeply digs the information of the super visualization of the underlying image bymeans of computer-aided engineering method. Then we will respectively con-struct texture feature of lymph nodes in PET images and multi-resolution histo-gram feature of lymph nodes in CT images, and by combining them, a PET/CTcomposite eigenvector of function and morphology for the enlarged lymph nodesin the non-small cell lung cancer is formed.Against the classification and recognition of metastatic lymph nodes inPET/CT images, this paper constructs a SVM classifier based on PET/CT compo-site eigenvector of function and morphology and perform the classification andrecognition of SUV concentrated metastatic lymph nodes in PET/CT images bymeans of the superior performance for SVM in terms of the ability of processing high dimension and small sample generalization. Respectively make the texturefeature of enlarged lymph nodes in PET images, the multi-resolution histogramfeature of enlarged lymph nodes in CT images and the PET/CT composite featureof function and morphology as the input feature of SVM. By training and testingthe sample set using SVM classifier, comparison and effective screening of thethree features are achieved. Experimental results show that based on the texturefeature and the multi-resolution histogram, the accuracy rate of the SVM classifi-cation and recognition of the PET/CT composite feature of function and mor-phology can reach86%which is higher than the classification results of the othertwo feature as the input feature, thus it can be used for auxiliary diagnosis ofconcentrated metastatic lymph nodes. |
