Abstract

Mixed dentition analysis has traditionally been performed using a linear regression for- 15 mula. However, population variation and lack of validity have been widely reported in such works. 16 In search of a solution, a size prediction model based on artificial neural networks (ANN) was de- 17 veloped. Various backpropagation-type ANN configurations were analysed. The upper and lower 18 incisors and upper molars were used as input variables, based on the previous study by Lara et al. 19 Different ANN structures are trained using a historical series of 200 plaster models for both training 20 and prediction. The results obtained show a good performance of the model over the 10 thousand 21 epochs, achieving a prediction accuracy between 0.75 and 0.99 measured by means of the mean 22 square error. Of the three models studied, the best performance with adequate use of computational 23 resources was 64 neurons, with three layers for the maxilla (r2=0.99) and four for mandible 24 (r2=0.904). It is concluded that ANNs allow a prediction of dental size that can be effectively used 25 for the analysis of mixed dentition in children from 7 to 12 years of age, being useful for pediatric 26 dentistry and orthodontics, rather Tanaka-Johnston method.