Research On Feature Extraction And Classification Algorithm Of Medical Image Based On NSCT

Breast cancer is the most common form of cancer in women.The diagnosis and treatment of breast cancer patients are generally performed through mammograms because the features of high-resolution and prominent breast details possessed by mammograms can improve the diagnostic accuracy.However,due to technical limitations,when shooting an X-ray image,it will inevitably introduce noise into the image,which will affect the doctors' analysis and judgment,and it is more likely to lead to clinical misdiagnosis and irreversible effects on the patient's diagnosis and treatment.Therefore,pretreatment of mammograms is an indispensable step in the diagnosis of patients with dense breast cancer.And a method that can automatically and efficiently preprocess mammograms has become a hot research topic in this field.As an important approach to extracting image information,feature extraction methods have been widely used in face recognition,image detection and other related fields.Wavelet transform is a common feature extraction method for mammograms,but it will lose a lot of useful information in the process of feature extraction,resulting in frequent occurrence of misdiagnosis.Classification is an important part of machine learning and data mining applications.Moreover the quality of the data determines the classification effect and time complexity of the classification algorithm,so data cleaning is a necessary step of the classification operation.In addition,the learning ability of a single classification model is limited,so the multi-model fusion classification method should be considered to improve the prediction and generalization ability of the model.According to the above questions,the main research content of this paper is as follows:(1)Aiming at the pretreatment of dense breast mammograms,this paper proposes a method based on mathematical morphology.The method based on mathematical morphology has the characteristics of nonlinear filtering,so it can not only alleviate the problem of lower image sensitivity due to dense breast,but also can effectively remove the invalid noise caused during image shooting.(2)For pretreated mammograms,this paper proposes a feature extraction method based on non-subsampled contour transformation(NSCT).Firstly,through NSCT image decomposition,a series of characteristic subgraphs of the original mammograms are obtained.Then feature data are extracted from the feature subgraphs by Zernike moments,that is,it takes the feature subgraph obtained through NSCT as input,and outputs high-order Zernike matrix,which preserves the feature information of the original mammogram.(3)For the data extracted by NSCT feature,this paper proposes a combined classifier model.First,in the feature selection stage,the features of the original mammogram can be better selected by using the model selection method.Then,in the stage of feature dimensionality reduction,the principal features analysis(PCA)is used to remove the secondary features and preserve the main features.Reduce the amount of feature data while improving the quality of feature data.Finally,the combined classifier model is designed by a gradient descent method,the prepared feature data is sent to the model for learning and the final classification model is obtained.