Research On Recognition Of Liver Cirrhosis Based On Deep Learning Algorithm

Cirrhosis is an irreversible chronic disease,which has important clinical significance for the diagnosis of cirrhosis.The method of clinician’s visual diagnosis of liver ultrasound images is subjective and not conductive to the diagnosis and recognition of lesions.Computer-aided diagnosis techniques using machine learning or deep learning methods provide effective intelligent technical support for the study of liver lesions.The manual extraction method of machine learning has weak ability of feature representation.Therefore,this paper first fuses HOG and equivalent LBP features in parallel,and recognizes them by LS＿SVM classifier,which achieves a better classification effect than other single feature recognition.In contrast,there are still problems such as low recognition rate and weak feature representation ability.In response to the above problems,this paper adopts deep learning theory and proposes two cirrhosis recognition methods to enhance the ability to express the characteristics of cirrhosis and improve the recognition rate of cirrhosis diseases.The main work of this article is as follows:(1)A cirrhosis recognition method using Inception v1 structure combined with VGGNet is proposed.First,optimize the VGGNet network parameters to adapt to the sample scale;second,use the Inception v1 structure to replace the relevant convolutional layer group of the VGGNet network.After the replacement VGGNet network,through multiple different scale convolutions of the Inception v1 structure,both can be extracted to more complete cirrhosis texture feature information,the depth and width of the VGGNet network can also be efficiently expanded;in addition,the network uses the 1×1 convolution of the Inception v1 structure,which greatly reduces the amount of network parameters and improves the network performance.Experiments show that the highest recognition rate of VGGNet combined with Inception v1 can reach 99.2%.(2)This paper proposes a cirrhosis recognition method based on spatial transformation network and heterogeneous convolution filter(ie SH＿ImAlexNet method).First,based on ImAlexNet(improved AlexNet),this paper improves the spatial invariance of the network by fusing the spatial transformation network(STN)to enhance the extraction ability of liver cirrhosis features and improve the classification and recognition rate of the network;second,in the above based on the fusion of heterogeneous convolution filters,not only the complexity of the network model is optimized,but also the recognition rate and operation efficiency of the network are further improved.Experiments show that the method in this paper can improve the effectiveness of the model to a certain extent while ensuring the recognition rate of cirrhosis.Compared with the traditional machine learning method,the deep learning method proposed in this paper is more suitable for the diagnosis and recognition of liver cirrhosis features,which helps to reduce the misjudgment of physician’s subjective differences and has more practical significance for the study of liver cirrhosis recognition.