Effectiveness of Machine Learning in Predicting Orthodontic Tooth Extractions: A Multi-Institutional Study.

Area of Science:

• Dentistry
• Artificial Intelligence
• Medical Informatics

Background:

• Orthodontic treatment planning often involves complex decisions regarding tooth extraction.
• Predicting the need for extraction versus non-extraction treatment can significantly impact treatment outcomes and duration.
• Leveraging machine learning offers a potential avenue for objective and data-driven treatment planning.

Purpose of the Study:

• To evaluate the efficacy of machine learning models in predicting the necessity of extraction versus non-extraction orthodontic treatment.
• To compare the performance of models trained on data from two distinct university institutions.
• To identify key clinical and cephalometric features influencing extraction decisions.

Main Methods:

• Utilized datasets from two universities, comprising 1135 orthodontic patients.
• Developed and applied Random Forest (RF) models using 20 input features (9 clinical, 11 cephalometric).
• Assessed model performance using sensitivity, specificity, accuracy, and feature ranking; performed cross-prediction between datasets.

Main Results:

• The combined dataset model achieved the highest performance: 50% sensitivity, 97% specificity, and 85% accuracy.
• Cross-prediction between University 1 and University 2 models resulted in a performance decrease of 0%–20%.
• Maxillary and mandibular crowding were identified as the most influential features for extraction decisions.

Conclusions:

• Machine learning, particularly RF models, demonstrates significant potential in aiding orthodontic treatment planning for extraction decisions.
• Models trained on combined data show robust predictive capabilities, though cross-institutional application requires careful validation.
• Maxillary and mandibular crowding are critical factors that AI models can effectively utilize for predicting treatment pathways.