Comparison the predicting effect of colonic adenomas high grade intraepithelial neoplasia by support vector machine and logistic regression analysis

Background and Aims: Logistic regression analysis is often limited by insufficient cases number.We set up predicting model by support vector machine(SVM)and logistic regression to predict the high grade intraepithelial neoplasia (HGIN)of colonic adenomas,and compared the results between two methods.
Patients and Methods: SVM is a mathematical classifying technique based on the principle of structural risk minimization.SVM can obtain the optimum result from limited samples. 157 patients with colon adenomas examined in our department from 2008 to 2010 were included and 12 clinical and endoscopic data were collected.50 randomized patients selected as training set to build a SVM model,by this model,we predicted the occurrence of HGIN in three testing sets with known pathology acquired by endoscopic resection.In the meantime,we built a logistic regression model used the data of total 157 patients and compared the predicting effect with SVM.
Results: The average predicting accuracy,sensitivity,specificity of SVM model in three testing sets were 92.6±3.3%,80.6±17.3%,94.8±0.6%. The predicting accuracy of logistic regression was 90.4%.The results between two models were similar.
Conclusion: SVM achieved satisfactory effect in predicting the occurrence of HGIN in colonic adenoma and compared to logistic regression only a small number samples were needed.