Development and validation of an artificial intelligence-based multimodal model for predicting orthodontic treatment duration and appliance type

Abstract

Precise estimation of the treatment duration and selection of proper appliances have always been a challenge clinically due to individual variables such as growth, compliance, and complexity of the case. At present, advances in artificial intelligence (AI) provide opportunities for improvement in diagnostic precision by integrating clinical and imaging data. This research assessed the development and validation of an ensemble learning-based AI model that could estimate with accuracy the duration of orthodontic treatment by integrating cephalometric measurements with standardized photographic data for the determination of appliance type between fixed appliances and clear aligners. We conducted a retrospective observational research study using anonymized data of 450 orthodontic patients aged between 12 and 35 years. Each record included pre-treatment lateral cephalograms, intraoral and extraoral photographs, appliance type, and treatment duration. The dataset was further divided into two parts – an 80% training set and a 20% testing set – in a random manner. Four models have been trained and compared: Support vector machines, decision trees, convolutional neural networks, and ensemble (hybrid fusion of cephalometric and photographic features). Predictive performance was assessed for accuracy, precision, recall, F1 score, and area under the receiver operating characteristic (AUC-ROC). Model predictions were compared with expert consensus using Cohen’s κ statistic and Bland–Altman analysis. The ensemble model performed best, with an accuracy of 89%, precision of 90%, recall of 88%, F1 score of 0.89, and AUC-ROC of 0.92. For treatment duration, the mean absolute error was 1.5 months (95% confidence interval: 1.2–1.8), which reflects about 7–8% deviation from the average clinical duration of 20.7 ± 4.9 months. Model–expert agreement was highly reliable (κ = 0.86), demonstrating clinically acceptable error margins. An ensemble AI model that integrates cephalometric and photographic features reliably predicts the duration of orthodontic treatment and appliance type, and perhaps in the future, an adjunctive tool for evidence-based treatment planning.