Building and Evaluating an Orthodontic Natural Language Processing Model for Automated Clinical Note Information Extraction

Area of Science:

• Biomedical Informatics
• Dental Research
• Machine Learning in Healthcare

Background:

• Malocclusion poses functional and aesthetic challenges, with treatment planning often subjective.
• Electronic dental records (EDRs) contain valuable data, but free-text notes hinder analysis.
• Objective, data-driven approaches are needed to standardize orthodontic diagnosis and treatment.

Purpose of the Study:

• To develop an Orthodontic Natural Language Processing (ONLP) model for structured data extraction from EDRs.
• To identify critical features influencing malocclusion using machine learning (ML).
• To enhance the objectivity and consistency of orthodontic treatment planning.

Main Methods:

• Utilized supervised (NER) and unsupervised (K-means) approaches for the ONLP model.
• Trained and validated ONLP and ML models on data from 7693 orthodontic patients.
• Applied ML models (Logistic Regression, Random Forest, XGBoost) to determine feature importance.

Main Results:

• The ONLP model achieved high accuracy (91%) in extracting orthodontic information.
• Supervised models showed 84% accuracy, excelling in Class I and III malocclusion identification.
• Key features identified include crowding, overjet, arch perimeter, spacing, crossbite, midline deviation, and occlusal wear.

Conclusions:

• The ONLP model offers a novel method for automating orthodontic data extraction.
• Enables advanced big data analytics for orthodontic research.
• Facilitates data-driven improvements in orthodontic care and research.