[Application of radial basis function neural network for grading of gliomas]

This retrospective investigation was directed to the applicability of Radial Basis Function Neural Network (RBF-NN) and Discriminant Analysis in the grading of gliomas. The data on 116 patients with primary glioma in our hospital from February 2008 to April 2009 were collected. Kruskal-Wallis H test was used to draw in the variable age ranks and then to take them out from the range of different grades of gliomas. The results of RBF-NN model, discriminant analysis, and the combined model of RBF-NN and discriminant analysis were evaluated and compared respectively with and without age. In this study, different classifications of gliomas showed statistically significant differences in age: and the accuracy of the models with age was better than the ones without age. The predictive accuracy and Kappa value of RBF-NN model and the combined model were also better than those exhibited by Bayes discriminant analysis. Consequently, as a prediction model, or to help other models, RBF-NN is of significance to predicting the grade of gliomas.