The Research On Tumor Diagnosis Algorithm Based On Neural Network

At present,malignant tumors have become one of the main causes of death in the world.It not only seriously affects human health and endangers the lives,moreover,its incidence has been increasing year by year,and in recent years,it has gradually shown a trend of younger ages.So far,no good tumor prediction method has been found.Therefore,the diagnosis and treatment of tumors have become the top priority of medical research.However,the diagnosis of tumors is complicated and the influencing factors are diverse,how to improve the diagnosis rate of tumors has become an urgent problem to be solved.Traditional tumor diagnosis methods include medical imaging,expert judgment,etc,but these methods rely more on the doctor’s subjective judgment,and the diagnostic accuracy depends on the doctor’s professional skills and clinical experience.In recent years,the rapid development of artificial intelligence technology has made artificial neural networks with strong adaptability and provided new methods for tumor diagnosis.Based on the characteristics of artificial neural network,this thesis applies neural network to research and experiment on tumor diagnosis methods,and compares and analyzes the research results.The main research contents of this thesis are as follows:(1)Diagnosis algorithm of breast tumor based on improved LVQ neural network.The Learning Vector Quantization Neural Network(LVQ)has a high degree of nonlinear computing ability,clustering of similar features,large-scale parallel processing and strong fault tolerance.However,its accuracy of tumor diagnosis is affected by the number of nodes in the competitive layer and the learning rate.In response to these shortcomings,an improved LVQ neural network algorithm model for tumor diagnosis is proposed.The experimental results show that the diagnostic accuracy rate of the proposed improved method on the Wisconsin Breast Cancer Database is 94.2%,which is 2.9% lower than the misdiagnosis rate of neural network.(2)NRS-GA based compound LVQ tumor diagnosis algorithm.In the tumor diagnosis algorithm based on improved LVQ neural network,it is found that LVQ neural network is sensitive to initial weights and the uncertainty of data leads to low diagnostic accuracy.Therefore,a compound LVQ neural network algorithm model based on neighborhood rough set and genetic algorithm optimization is proposed.The neighborhood rough set effectively removes the redund ant features of the data and speeds up the convergence of the neural network;the genetic algorithm is used to optimize the initial weight of the neural network,and the nonlinear mapping relationship between breast tumors and symptoms is constructed by using the proposed algorithm model.Similarly,the data of breast tumor in the UCI machine learning database are the test sample,the results show that the proposed algorithm is effective,which has an average diagnostic accuracy of 97.1% a nd a maximum of98.55%.(3)tumor diagnosis algorithm based on convolutional migration network.Aiming at the problem of the low diagnosis rate in computer-aided tumor diagnosis based on automatic multiple classificat ion of histopathological images,a tumor diagnosis algorithm model combining convolutional network(CNN)and Transfer learning(TL)is proposed.After the image is preprocessed,the nuclear extraction scheme is used to extract non-overlapping nuclear,and then the convolutional neural network and nuclear guide learning method are combined to extract feature vectors from the non-overlapping core blocks,and the 3-norm is introduced to the feature fusion stage,finally,gradient enhanced tree classifier is used to classify the fusion feature.The experimental results show on the public Brea KHis database,the accuracy rates of 40 X,100X,200 X and 400 X are 94.12%,93.49%,94.73%and 93.27%,respectively.