The Study Of Application With Logistic Regression In Comprehensive Analysis And Diagnosis Of Breast Cancer

ObjectiveThe aim of the study is to establish a logistic aided diagnosis ofbreast cancer model,through the comprehensive analysis of logisticalregression with the basic situation,clinical examination, mammographyx-ray, color Doppler imaging studies four types information about breastpatients.We try to use the model created by integrating multiple data toredict the breat mass benign or malignant.The model can provide help inthe identification of benign or malignant breast disease and improve thediagnostic accuracy of breast disease.MethodsObject of study: Collect breast patients of Nanhua Hospital and FirstAffiliated Hospital from2009to2011.Request of the selected cases:1.Cases must be confirmed by biopsy.2.The information of the basicsituation、symptoms and signs、Mammography、Ultrasound must becomplete.3. The Collection of cases must be objective, accurate, completeand correct. Iuput the datas：First,put the datas into Microsoft Excel2003, and then imported them into SPSS17.0software. All the following analy sis are used the SPSS17.0Software. Univariate analysis:We can get thestatistically significant variables through univariate analysis of allvariables. Multivariate analysis:Put all statistically significant variablesinto logistic regress analysis,we can get the final variables that used toestablish models. Establish diagnostic models: We can use differentcombinations of image data as a variable establishing different diagnosismodel. Such as: we used the basic information（B）, symptoms andsigns（S）, mammography x-ray （M） and color doppler ultrasound（U） as avariable establishing a model, put them into multi-factor logistic regressanalysis,we can get BSMU-logistic model. we used the basicinformation（B）. symptoms and signs（S） and mammography x-ray（M） asa variable establishing a model, we can get BSM-logistic model.. we usedthe basic information（B）,symptoms and signs（S） and color dopplerultrasound （U）as a variable establishing a model, we can get BSU-logisticmodel.Comparison of models:We use ROC curve and medcal Softwarecompare the diagnostic accuracy of the three models.Select the optimalmodel and its best diagnostic cut-off point:we choose the optimal odelaccording to the area under the ROC curve,and select the best diagnosticcut-off point as the diagnostic criteria of breast cancer.Results 1.Univariate analysis: Age, occupation, lesion texture, lesionboundary, lesion morphology, surface smoothness, activity, axillarylymph nodes, edge of mammography, shape of mammography,microcalcifications of mammography,tructural disorder, thick levy, funnel levy, edgeof ultrasound, morphology of ultrasound, microcalcifications ofultrasound, peripheral and internal flow signals, flow signalsclassification （CDFI）, ultrasound axillary lymph nodes in the case groupand control group are significant difference（p <0.05）.2.Multi-factoranalysis: put all statistically significant variables into logistic regressanalysis,we get the final variables,such as: lesions texture, lesionboundary, edge of mammography,funnel levy,grape of ultrasound,peripheral and internal flow signals, flow signals classification （CDFI）,ultrasound axillary lymph nodes.3.The diagnostic accuracy of threemodels: The diagnostic accuracy of BSMU-logistic model, BSM-logisticmodel,BSU-logistic model are respectively93.4%,85.6%,87.4%, thespecificity88.0%,74.0%,70.0%, and the sensitivity95.7%,90.6%,94.9%.4.Comparison of models: The under the ROC curve area （AUC1、AUC2、AUC3） of the three models were0.964,0.924,0.920. The results ofthe comparison between the two groups: AUC1and AUC2, AUC1andAUC3had statistically significant, but AUC2and AUC3hadn’t statisticallysignificant.5. Select the optimal model and its best diagn-ostic cut-offpoint:ROC area under the curve suggests that BSMU-logistic model is the optimal model. The regression equation of the model is: Logit（P）=-48.335+2.227*X₁+1.033*X₂+2.684*X₃+21.633*X₄+2.417*X₅+8.101*X₆-2.826*X₇+2.527*X₈.And the best diagnostic cut-off point of themodel is0.501, which corresponds to the sensitivity of95.7%and aspecificity of88.0%.Conclusion1. Using multiple data to establish the logistic regression of breastcancer diagnostic model can improve the diagnostic accuracy of breastcancer.2. Logistic regression model of mammography and ultrasound hadnot significant difference in the accuracy of the diagnosis of breastcancer.