Caries Prediction Using Conventional Methods and Artificial Intelligence Neural Network

Introduction: Artificial intelligence neural network (ANN) has been applied in health sciences with promising results in disease diagnosis, prognosis and prediction. However, its potential remains unexplored in dentistry. Objective: to compare the capability of conventional statistical methods (CSM) and ANN in constructing caries risk assessment/prediction models. Methods: 1782 preschoolers (3-5 years of age) in Singapore participated in a questionnaire survey (completed by parents), with an oral examination and salivary tests on stimulated saliva flow rate, buffering capacity, level of Streptococcus mutans, level of Lactobacilli, and plaque pH. The caries status of preschoolers was assessed at baseline and after 12 months. Caries risk prediction models were constructed with CSM (multiple logistic, ordinal, and linear regressions) and ANN (C4.5, SVM, Nbay, and Multi Layer Perceptron). Results: ANN predicted the caries increment more accurately when limited information was available, such as “only behavioral factors”, “demographic factors + oral hygiene status”, or “demographic factors + biological tests” (all p<0.05). As compared with linear regression, Multi Layer Perceptron reached a higher accuracy in predicting “number of new affected surfaces” (p<0.05). Conclusion: ANN is promising in the development of caries prediction models with a greater simplicity and accuracy when compared with conventional statistical methods. (This study was financially supported by National University of Singapore Academic Research Funds R222-000-021-112 and R222-000-022-112.)