Intelligent Diagnosis Method Of Echinococcosis In Liver Ultrasound Images

Echinococcosis is a parasitic disease caused by human infection with Echinococcus.Accurate diagnosis and classification of echinococcosis is necessary for subsequent treatment of the disease.Patients with echinococcosis are widely distributed throughout the world,especially pastoral areas and semi-agricultural and semi-pastoral areas in western China.Because the pastoral area is sparsely populated and there are insufficient professional doctors,patients with echinococcosis cannot get timely diagnosis and effective treatment.Ultrasound is the most commonly used tool for clinical diagnosis of liver echinococcosis.However,screening for large areas of liver echinococcosis is time-consuming and labor-intensive,and the diagnosis of liver hydatid disease is very much dependent on the personal experience of the doctor.Therefore,in order to improve the efficiency of screening and diagnosis of liver echinococcosis and alleviate the pressure of insufficient doctors in pastoral areas,this paper proposes an intelligent assisted diagnosis method for liver echinococcosis based on liver ultrasound images.First,this paper proposes a two-stage model from coarse to fine to realize the intelligent assistant diagnosis of liver echinococcosis.In the first stage,we consider the problem of automatic detection of liver echinococcosis as the lesions location and group classification task,and implement an echinococcosis detection network(Echinococcosis Detection Networks,EDnet)based on deep convolutional neural networks,to detect the lesions of liver echinococcosis and make a group classification.In the second stage,we implement fine-grained classification of echinococcosis lesions on the basis of group classification,and implement an echinococcosis classification network(Echinococcosis Classification Networks,ECnet)based on fine-grained visual classification,to classify the lesions of liver echinococcosis.In addition,this paper also proposes an intelligent auxiliary diagnosis method of echinococcosis based on image generation.We discussed the ultrasound image generation method for echinococcosis by Generative Adversarial Network(GAN)to solve the problem of imbalanced data of liver echinococcosis ultrasound image and improve the accuracy of classification.In this method,we build a multi-scale conditional generation adversarial network and use negative ultrasound images of the liver to generate ultrasound images of liver echinococcosis.This method can generate high-resolution ultrasound images of liver echinococcosis,while maintaining the original structure and image texture of the ultrasound image.In the same time,this method can improve the accuracy of the deep learning model for classification of liver echinococcosis.The experiment proves that the two-stage method from coarse to fine proposed in this paper can effectively overcome the problems of large intra-class and small interclass differences in ultrasound images of liver echinococcosis,solve the problem of the echinococcosis lesions vary so much in size,and the problem of fuzzy boundary of alveolar echinococcosis lesions.Our method has better classification performance than popular deep learning methods.The average recall,accuracy and F1 score reached 87.1%,86.2% and 86.5%,respectively.In addition,the method based on image generation can solve the problem of data imbalance of echinococcosis.Finally,the average F1 scores increased respectively by 4.1% and 1.3% for CE2 and AE3,which types have less data.At the same time,this paper is the first intelligent auxiliary diagnosis system for liver echinococcosis in ultrasound images,which can be used for clinical screening and diagnosis of liver echinococcosis,and has good application prospects.