| Gastric cancer is a worldwide common malignant tumor，its morbidity isthe highest in the various types of tumors. According to statistics,there areabout647000people died of stomach cancer every year.In our country, themortality of gastric cancer is only after lung cancer and liver cancer,it’s aserious threat to human life and health. Gastric cancer is not easy to be found,the symptoms and signs are hidden, so advanced gastric cancer is the mostcommon type in clinical gastric cancers. China’s advanced gastric canceraccounts for about90%of all gastric cancer cases. The most commonmetastatic pathway is the lymph node metastasis, lymph node metastasis is theimportant factor of cancer recurrence and death. The preoperative evaluation oflymph node metastasis in gastric carcinoma is helpful for selecting thereasonable range of lymphadenectomy,and is also important for the judgementof prognosis. The preoperative assessment method for lymph node metastasisin gastric carcinoma includes imaging examination,just like CT, endoscopicultrasonography (EUS),nuclear magnetic resonance imaging (MRI), positronemission tomography (PET), and there also include molecular markers, thesentinel lymph node (SLN), computer technology and so on,but these methodsall have shortcomings. This study uses multi-layer perceptron artificial neuralnetwork (MLP) and Logistic two regression analysis method for preoperativeprediction of lymph node metastasis in patients with advanced gastric cancer, inorder to explore better preoperative prediction method.Objective: According to the gender、age、smoking、drinking history、medical history、first symptom、position、depth of invasion、pathological typeand Borrmann type of advanced gastric cancer patients，use MPT and Logisticregression analysis to predict the occurrence of lymph node metastasis. In addition, compare the accuracy of the two statistical methods and find aneffective method for preoperative prediction of lymph node metastasis inadvanced gastric cancer patients.Methods: from700patients who had operated and the pathologicaldiagnosis for advanced gastric cancer in period from December,2003to January2011in the JiLin University China Japan Union Hospital,we assemble theinformation of these patients,including gender、age、smoking、drinking history、medical history、first symptom、position、depth of invasion、pathological type、Borrmann type and wheather the Lymph node metastasis or not. We usemultilayer perceptron artificial neural network and logistic two regressionanalysis method for the preoperative data analysis, statistics(this step iscompleted by SPSS17.0statistical software packages.)， and draw the receiveroperating characteristic (ROC) curve, according to the area under the curvecompare the two statistical methods with the predictive ability in patients withadvanced gastric cancer for lymph node metastasis.Results: The result of χ2test is Gastroscope Borrmann type, tumorposition, medical history and depth of invasion are “risk variables”, thedifferences between the4factors for lymph node metastasis are statisticallysignificant. The prediction accuracy of ANN1(established on the4" riskvariables") is70.4%, the square measure of area under the ROC curve is0.796.The prediction accuracy of ANN2(established on all the10variables) is75%,the square measure of area under the ROC curve is0.831. Logistic two elementregression analysis showed, the variables which have significant correlationwith lymph node metastasis in advanced gastric cancer are(according to thecorrelation from big to small): the depth of invasion、endoscopic Borrmanntype、pathological type、drinking history and the tumor position.We put all the10variables into the SPSS17.0software packages,the Logistic regressionanalysis system automatically put the above5variables into the Logistic regression equation, The prediction accuracy of group Logistic is69.3%, thesquare measure of area under the ROC curve is0.771.Conclusion:1.The lymph node metastasis of advanced gastric cancer issignificantly correlated with invasion depth, position, endoscopic Borrmanntype, drinking history and pathological type.2. The more independent variables, the more samples, the more intensiveartificial neural network, the accuracy and specificity of MLP are higher.3.When the independent variables are less, the artificial neural network isnot superior to Logistic regression analysis; when the independent variables aremany,the prediction accuracy of artificial neural network is higher than that ofLogistic regression analysis.4. The binding of artificial neural network and clinical information can beused for preoperative prediction of lymph node metastasis in advanced gastriccancer, and provide guidance for operation mode selection, postoperativetreatment options and prognosis judgment. |
PDF Full Text Request