The Study Of Low Rectal Cancer Prognostic Model Based On Artificial Neural Network Data Mining

Objective: To build a Five-year surviving condition predicting model for low rectal cancer patients who had completed the operations.The model utilized collected clinical data that covered various prognostic factors with the application of Artificial Neural Network(ANN)data mining technology.In the meantime,to reflect on and to evaluate the model's quality of performance with the comparison of the traditional linear statistical analytical model.Methods: Analyze the collected clinical data of 186 low rectal cancer patients who received surgical operation at the Affiliated Hospital of Qingdao University from January 2009 to August 2011.Randomly divide the patients into two test sets: the training set,a group of 150 cases,in order to accomplish the data mining process used to build the cancer prognostic model;the test set,a group of 36 cases,that was not used in the data mining process,in order to evaluate the effectiveness of the prognostic model.The data mining method used in this process was the ANN.Results: The Five-year surviving condition of low rectal cancer patients is related to 7 indicators: T Stage,maximum tumor diameter,lymphatic metastasis,distant metastasis,surgical procedure,serum CEA level,and pathological type(P<0.005).The ANN predicting model of the Five-year surviving condition of the patients had an accuracy of 86.11%,a sensitivity of 75.00%,and a specificity 89.29%.The Logistic Regression predicting model had an accuracy of 77.78%,a sensitivity of 55.56% and a specificity of 85.19%.In general,ANN model performed better than Logistic regression model.Conclusions: The ANN data mining technology is able to discover significant prognostic factors from the complex clinical data of low rectal cancer patients and to form effective predicting model that examines the living condition of patients 5 years after the operation.