Recognition Of Liver Cirrhosis Based On Convolutional Neural Network

Cirrhosis is a common disease in people's digestive system.Many causes can cause cirrhosis and threaten people's lives and health.Therefore,it is of great practical significance to study the identification of cirrhosis.However,because clinicians have certain subjective errors in the diagnosis of liver diseases based on medical images,and liver images have unclear pathological lines and uneven texture features.Therefore,the use of computer-aided diagnosis technology for the diagnosis of liver cirrhosis helps to improve the efficiency and accuracy of clinical diagnosis.Traditional machine learning uses a variety of feature extraction algorithms to form feature vectors and input the feature vectors into the classifier for classification.The accuracy of diagnosis and recognition is not high,and there is a lot of waste of time and manpower.Therefore,traditional machine learning cannot meet the needs of diagnosis.For the above shortcomings,the deep learning algorithm is used later in this article.Its automatic learning of more distinguishing features and recognition and classification features solves the above problems well and improves the accuracy of identifying and distinguishing liver cirrhosis from normal liver pathology rate.This paper mainly proposes two deep learning recognition algorithms:(1)A multi-scale and multi-feature CNN model based on data enhancement is proposed.Firstly,data enhancement was carried out on the limited data set of cirrhosis,and samples of three different scales were used as input to the model.The model learned the characteristics of different scales at the same time.Then,the weighted sum of the multiscale information of different layers of the network is done.In addition,the weight coefficient of the classifier is improved,the weight sensitivity is adjusted flexibly,and the fusion of features at all scales is more conducive to capturing the varied texture information in the cirrhosis image through experiments,which finally improves the performance of the whole algorithm,and the experimental accuracy reaches 99.2%.(2)A method combining the lightweight model structure MobileNet V2 and ELM was proposed to identify cirrhosis.With the method of transfer learning,pre-training was conducted on the sample graph of 25,000 dog and cat data sets,and the weight and parameters obtained were used as initialization parameters to avoid the occurrence of overfitting.In order to improve the accuracy of recognizing cirrhosis,the full-connection layer features of the model were output in vector form and sent to ELM for classification,replacing the softmax classifier.The test time of the final experiment is lower than that of other models,which can be as low as 12.7 seconds.Compared to the traditional feature extraction methods,the deep learning model is more suitable for actual clinical medical identification diagnosis,both to shorten the time of the patients diagnosis of disease,and improve the accuracy,greatly reduces the clinical doctor degree of the influence of subjective factors,has the very good practical significance.