Design And Implementation Of Liver Disease Diagnosis Model Based On Autoencoder

Liver disease seriously threatens the health of modern people because of its high infectivity and high incidence.The emergence of deep learning has greatly promoted the development of liver disease diagnosis models.However,the high dimensionality and high noise of liver disease data have always been the main reasons affecting the accuracy and efficiency of the model.Therefore,how to use deep learning to extract important feature information from complex liver disease data and establish an accurate and efficient liver disease diagnosis model is one of the hot spots in liver disease diagnosis research in recent years.This study takes liver disease diagnosis as the main research object,and discusses the feature extraction method and the establishment of the diagnosis model in the process of establishing the diagnosis model.First,based on the improved Autoencoder(AE)feature extraction of liver disease data;Then establish an Adaptive Integrated Stacked Correlation Denoising Autoencoder liver disease diagnosis model(SISCDA);Finally design and implement a liver disease auxiliary diagnosis system.The specific work is as follows:(1)A new feature extraction algorithm SCA is proposed to solve the problem that AE ignores the change of data structure relationship in the reconstruction process during feature extraction,resulting in insufficient feature extraction capability.The algorithm keeps the feature data relationship structure before and after reconstruction unchanged by adding global and local association limits to the AE’s objective function,and calls the AE with added association limits as CAE.Then multiple CAEs are stacked into a stack-associated self-encoder(SCA)to further improve the feature extraction capability of the algorithm.The experimental results on 4 sets of experimental data sets show that the performance of SCA feature extraction algorithms are better than other machine learning feature processing algorithms and traditional AE algorithms.(2)Aiming at the problem that the accuracy of the diagnosis model is reduced by multiple noise interferences of liver disease data,this thesis proposes a new liver disease diagnosis model SISCDA based on noise reduction autoencoder(DAE)and SCA combined with integrated learning.The model first uses the denoising method of DAE to train the denoising of the SCA,so that the SCA has both the ability of feature extraction and denoising,and it is called the stack associated denoising autoencoder(SCDA).Then use integrated learning to integrate multiple SCDA to deal with the different noises of liver disease data.Aiming at the integration method,this thesis proposes an adaptive weight integration method.This method dynamically adjusts the weight according to the error of the two adjacent training results when updating the weight of SCDA.Experimental results show that compared with other integrated liver disease diagnosis models,SISCDA can effectively solve the noise problem of liver disease data and improve the accuracy of diagnosis.(3)On the basis of the above research,this thesis also designs and implements an assisted diagnosis system for liver disease based on related technologies of Java EE platform.The thesis first specifically analyzes the related requirements of the system,then designs the system architecture,function modules and database in detail,and finally uses the existing technology to realize the related system functions such as model training,liver disease diagnosis and statistical results display.While it is convenient for patients with liver disease,it also verifies the feasibility of the algorithm proposed in the thesis.