The Application Of SVM Classification Model In The Auxiliary Diagnosis Of Cancer

Cancer is often accompanied by the malignant tumors, which are serious harmful tohuman health and life.With the rapid development of technology for cancer diagnosisand treatment, whether the tumor has occurred EGFR (Epidermal Growth FactorReceptor) mutation is becoming a very important diagnostic indicator, which hassignificant implications for the targeted therapy of clinical patientsThe researches of this paper are conducted to explore a new scheme for thediagnosis of cancer mutation. And the specific research direction is focus on thediagnosis of brain tumor mutation, through the cooperation with Chongqing Dapinghospital. The existing diagnosis methods need to perform brain surgery on patients, toget brain tissues for gene sequence detection, will bring additional burden and risks tothe patients. At the same time, the conventional approaches always ignore the potentiallinks between different types of cancer mutations, while patients may got variouscancers. And the classification results are also easily influenced by the noise in theclinical data. This paper proposes an approach to solve the problems, with the mainwork is as follows:①A diagnosis scheme based on fusion of multiple types of cancer mutations isproposed. Since the easier diagnosis, the lung cancer mutation is introduced as anindependent variable, combined with other clinical characteristics for the diagnosis ofbrain tumor mutation. The combine of the two different cancers will find the potentialcontact between. The chi-square test is used for significance analysis of the clinicalcharacteristics, and result show that the lung mutation(P<0.01) is closely associatedwith brain mutations which means significant difference. And SVM classificationalgorithm is used to construct a classification model of brain tumor mutation. In thispaper, the experiments about whether lung mutation should be introduced is performed,and the results are evaluated by sensitivity, accuracy, specificity and AUC. Finally, theintroduction of lung mutation makes the classification better.②Feature extraction of clinical data is added before building the classificationmodel in the diagnosis scheme, and the process can filter noise in order to improve theresult. In this paper, a variety of feature extraction algorithms are used for comparison,including LDA algorithm(Linear Discriminant Analysis), PCA algorithm(PrincipalComponent Analysis), FA algorithm(Factor Analysis). And finally find out the use of LDA can get the best classification and can greatly improve the sensitivity, accuracy,specificity and AUC.③Developed a brain tumor mutation diagnosis application system according tothe above scheme.The main achievement and contribution of this paper is the introduction of lungmutation to the classification model which is used to classify brain mutation.