Application Of Machine Learning In Cancer Diagnosis

Cancer has become a serious threat to human health that can not be ignored nowdays.The traditional methods of cancer diagnosis are cell morphology,histopathology and so on.With the success of machine learning in computer vision,natural language processing and speech recognition,it has become a new operational method to predict cancer.According to the main content of cancer detection,machine learning technology is applied to the field of intelligent medical care,and models are built for cancer susceptibility,survivability and recurrence,respectively,to assist doctors to make decisions on the above issues.The main research work can be described as follows:(1)To study the susceptibility of cervical cancer.Aiming at the problem of imbalanced classfication and invalid features filtering in the diagnosis of cervical cancer,this thesis proposes a method combining particle swarm optimization(pso)algorithm with the synthesis minority oversampling technique from the perspectives of data distribution,model accuracy and feature number.This method is used to study cervical cancer data with a variety of machine learning models in order to solve the problem of imbalanced classfication in the data set and filtering out invalid features.Synthesis minority oversampling technique can better simulate the data distribution.Particle swarm optimization not only considers the model accuracy,but also solves the problem of manual feature number selection.Compared with traditional model using the feature recursion elimination method and principle component analysis method,this method avoids manual feature number selection.Experiments show that this method is better than the traditional method and can improve the model performance effectively.(2)To study the survival of lung cancer.Aiming at the one-year survival rate of lung cancer patients,a classification model combining particle swarm optimization(pso)with LightGBM model is proposed in this thesis,which sloving the problem of mannual selection of features in the face of complicated lung cancer data in high dimension.In addition,genetic algorithm is used as a control group in this thesis,and a variety of machine learning models are used for modeling,and the performance of each machine learning model is compared.Compared with random forest,support vector machine,neural network and other models,LightGBM combined with particle swarm optimization algorithm improves the classification accuracy of survival time of lung cancer.(3)To study the recurrence of breast cancer.Firstly,the prediction models of breast cancer recurrence are constructed by using various machine learning algorithms,and the models are integrated by using a stacking network.Then,the improved stacking network is used.For a single network node,after the stacking model is constructed and the k submodel is completed,all the data is used for training again,so as to obtain a single model node with more data for training.Finally,the effects of particle swarm optimization are compared.Experiments show that the improved stacking network model can combine the advantages of each model,and its performance is improved compared with the traditional stacking network model.