Research On Liver CT Image Classification Algorithm Based On Machine Learning

Early diagnosis and treatment of liver disease is an important measure to improve the cure rate of patients.Computed Tomography(CT)technology is an important means of noninvasive assisted diagnosis in medical science.The combination of CT image and machine learning technology for the assisted diagnosis of liver diseases has the advantages of high speed and efficiency,and it has become a hot topic in medical research.In order to improve the accuracy of liver classification based on CT images,this thesis studies the method of machine learning for liver disease analysis.The specific research contents are as follows:(1)In view of the small amount of clinically labeled CT images with abnormal liver function and little difference,a new Neighborhood Average Local Ternary Direction Pattern,based on Neighborhood average local ternary direction Pattern,NALTDP(Liver CT image classification method).Firstly,liver CT images are preprocessed to extract normal and abnormal images of the same size.Then,the neighborhood mean is calculated by using the correlation of the center pixel’s neighborhood changes in the adjacent eight directions,and the neighborhood mean is encoded by the three-value pattern,the direction pattern and the three-value direction pattern.Finally,the histogram is classified and recognized by support vector machine.By extracting the neighborhood change information and direction information in each direction,NALTDP operator expresses more levels of local texture information and highlights the differences among liver function images.Experiments show that NALTDP has better classification effect than other methods.(2)Considering the convenience of computer-aided diagnosis by doctors and the urgency of patient diagnosis,lightweight convolution networks are adopted to shorten network complexity and reduce the number of training parameters for CT image classification of different liver diseases such as hepatic cyst and hepatic hemangioma.A Global attention Mobile Net network(Global attention Mobile Net,GA＿Mobile Net)is proposed on the basis of traditional lightweight networks.This method designs a lightweight global attention network based on Mobile Net V3-Large network architecture,focusing on focal region feature information.By introducing the global attention module,the global information feature of the focal area is enhanced,and the receptive field of the network is enlarged,so as to improve the feature expression ability of the network.The experimental results show that the GA＿Mobile Net network can significantly improve the classification performance of liver diseases through multiple evaluation indexes under the condition of low complexity of network model.(3)Aiming at the single feature information extracted by a single lightweight network in CT image classification studies of different liver diseases,a hybrid lightweight network classification and recognition method of liver CT images based on feature selection was proposed.In order to obtain multiple groups of feature information in the image,the method inputs the input images into three lightweight networks respectively to get three groups of feature images.In order to represent more feature information and reduce the computing cost of the network,feature selection method is used to select the most representative features.At the same time,the support vector machine is used to replace the full connection layer to improve the generalization performance of the network,so as to realize the classification and recognition of different kinds of liver disease CT images.The experimental results show that the hybrid lightweight network based on feature selection has strong classification and recognition ability in liver CT images on multiple evaluation indexes.