Skip to Main Content
Objective: To study the artificial neural network in the diagnosis of the smear negative pulmonary tuberculosis. Methods: All original data was randomized into modeling sample and validating sample. The modeling sample was further randomized into training sample and testing sample. The training sample was used to screen out significant single parameters and to develop the diagnostic model of smear negative pulmonary tuberculosis based on artificial neural networks. The testing sample was used to determine the appropriate architecture of the model. The validating sample was used to evaluate generalization of this model. Results: The architecture of artificial neural network is (29-9-1)-BP. When the model was applied to the validating sample, the area under the receiver operating characteristic curve was 0.989plusmn0.015, with accuracy, sensitivity and specificity at 93.10%, 88.89% and 100%, respectively. Conclusions: The artificial neural network model used in diagnosing smear negative pulmonary tuberculosis can be better generalized. As such, this can be used as a tool for the diagnosis of smear negative pulmonary tuberculosis and deserves further investigation.