Logistical Regression Model.Neural Network Model And Decision Tree Model In The Comparative Study Of The Imaging Diagnosis Of Breast Cancer

Objective：Establish color doppler ultrasound data of mammary disease diagnosismathematical model, to evaluate the diagnostic performance of the three model andcompare the various models,using mathematical model to assist the identification ofbenign and malignant in mammary disease, standardize the process of breastultrasound diagnosis and improve the diagnostic accuracy of breast masses.Methods：(1) Cases Collection:Collected The first affiliated hospital of University ofSouth China and the affiliated NanHua hospital firmed by surgical pathology biopsyof breast disease patients, total680cases, including the basic situation of patients,symptoms and signs, molybdenum target X-ray and (or) color doppler ultrasound andpathological results. For process this research, chose complete color doppler flowimages of552patients as the research object for this study.(2) Data entry: Input theresearch data by epidate3.1software, and then upload to SPSS17.0to establish theinformation database of patients with breast diseases （.3）Index selection: Using SPSSsoftware to filter the variables index, classify variables chi-square test, measurementdata were analyzed by t test, measurement data with P <0.05for the inspectionstandard.(4) Build Mathematical models：through the step screening index as themodeling of the indexes, via SPSS clementine12.0professional data mining softwaremaking artificial neural network model, the decision tree model, Logistical regressionmodel. the software will divide modeling process into two parts, one part is thetraining set, random sampling70%of the patients used to build the model; The otherpart is testing set, the remaining30%of the data used to test the performance andimprove the function which has been generated by the model.（5） calculate andcompare each model diagnosis on diagnosis accuracy, sensitivity, specific degrees. （6）Model diagnostic performance evaluation, drawing each model of the ROCcurve through medcal software, through the size of the area under the ROC curvemodel to compare the diagnostic performance.Results：(1)Modeling index screening results: it shows that the measurement dataof the ultrasound focal（16indexes）of statistically significance, could be used asmodeling index, they are： length to diameter, short diameter, blood flow signals ofcolour to exceed lesions of EDV, PVS, RI, the mubmer of mass,ultrasonicmorphology one, ultrasonic morphology two,envelope, microcalcification, internalecho one, internal echo two,peripheral and internal blood flow signal, axillary lymphnode enlargement and Blood flow signal classification.(2) Through analyze thetraining set and testing set to establish mathematical model, Diagnosis results of threemodels（3）Sensitivity, specific degrees and accuracy of three model: Logisticregression model was85.71%,60.66%and75.51%,respectively. Neural networkrespectively88.57%,55.74%and75.90%, the decision tree was94.29%、82.00%、89.76%, results indicat that the decision tree model have better sensitivity, specificdegree and the accuracy in diagnosis of breast cancer than other models.（4）Comparison between the three model, the area of three models under the ROC curve,Logistic regression model0.717,neural network model0.732, and the decision treemodel0.881, explain decision tree model has better performance in diagnostic.Conclusion：1. Using the computer data excavating technology to establishmathematical model of color doppler ultrasound data for the diagnosis of mammarycancer will help to improve the diagnosis accuracy.2. Compare the three mathematical models, the decision tree model ofbetter diagnostic accuracy and perfermance.