Prediction of unerupted canine and premolars width using artificial intelligence compared with Tanaka-Johnston and...

Accurate prediction of unerupted teeth widths significantly contributes to formulating optimal treatment plan during mixed dentition phase. The ability of artificial intelligence to identify hidden patterns and both linear and non-linear relationships can be particularly advantageous in this context.

In this retrospective observational study, pre-treatment dental casts of 336 orthodontic patients were scanned using a 3D scanner. The mesiodistal dimensions of each tooth were measured digitally, and the values for the left and right sides were averaged. These measurements, along with the patients' sex, were then used to train machine learning and regression models with various architectures. The dimensions of teeth #1, #2, #6 and a binary value representing patients' sex were used as input features to predict the sum of widths of teeth #3, #4, and #5 during the model training phase. The performance of the model was evaluated by calculating the mean absolute error (MAE) between the predicted values and the actual measurements obtained from the casts. Model performance was compared with Tanaka-Johnston equation and Moyers prediction tables.

Among the tested models, the Support Vector Machine showed the highest accuracy in predicting lower arch tooth sizes, achieving a MAE of 0.65mm and Standard Deviation (SD) of 0.82mm. For the upper arch, the Linear Regression model performed best with a MAE of 0.64mm and a SD of 0.80mm.

Our findings demonstrate that machine learning-based regression techniques offer superior accuracy in predicting the dimensions of unerupted teeth compared to traditional methods, making them a valuable tool for treatment planning in mixed dentition.