The classification model of type 2 diabetic peripheral neuropathy was studied based on GA-BP neural network model / 解放军医学杂志

ABSTRACT

Objective: To establish the genetic algorithm optimizing back-propagation (GA-BP) neural network model based on the clinical examination index for diagnosing type 2 diabetic peripheral neuropathy (DPN), and evaluate its diagnostic performance. Methods: A total of 2240 DPN patients and 2632 non-DPN patients were collected from the Hospital affiliated to Chongqing Medical University from January to December 2016, and univariate analysis was performed for 41 clinical test indicators of the two groups of patients with SPSS 21.0. Thirty-seven items of statistically significant variables were selected to establish the decision tree and Bayesian model with R software, MATLAB 2014a software was employed to establish the BP neural network and GA-BP neural network model, the advantages and disadvantages of these four models were compared with various evaluation parameters. Results: Using decision tree, the Bayesian, BP neural network and GA-BP neural network for 4872 cases of observation object model, the decision tree of The test sample accuracy with decision tree model was 93.4%, with Bayesian model was 70.0%, with BP neural network model was 98.9%, and with GA-BP neural network model was 99.5%. The areas under the ROC curve were 0.93, 0.72, 0.99 and 0.99, respectively. The Youden Indexes were 0.87, 0.59, 0.98 and 0.98, respectively. Conclusion: The GA-BP neural network established in present paper has a good computer-aided diagnosis function for type 2 diabetic peripheral neuropathy, but further clinical trials are still needed.