Classification of periodontitis stage and grade using natural language processing techniques.

Area of Science:

• Dentistry
• Artificial Intelligence
• Natural Language Processing

Background:

• Periodontitis is a complex inflammatory disease affecting dental tissues, often linked to microbiome imbalances.
• Early diagnosis of periodontitis is crucial for effective treatment and prevention of complications.
• Clinical Decision Support Systems (CDSS) offer potential for improving diagnostic accuracy and efficiency in healthcare.

Purpose of the Study:

• To develop and evaluate a CDSS utilizing Natural Language Processing (NLP) for early periodontitis diagnosis.
• To predict the stage and grade of periodontitis by extracting information from patient dental charts and clinician notes.
• To compare the performance of a BERT-based NLP model against a traditional feature-engineered model.

Main Methods:

• A CDSS was developed using the BERT (bidirectional encoder representation for transformers) model for NLP.
• The BERT model was fine-tuned on 70% of 309 anonymized patient periodontal charts and clinician notes.
• Performance was evaluated on 32 unseen patient records and compared to a baseline feature-engineered model with MLP techniques.

Main Results:

• The BERT-based CDSS achieved 77% accuracy in predicting periodontitis stage and 75% in predicting grade.
• The baseline MLP model achieved 59.4% accuracy for stage and 62.5% for grade on the same dataset.
• On unseen data, the BERT model showed higher accuracy (66% for stage, 72% for grade) compared to the baseline.

Conclusions:

• The BERT-based CDSS significantly outperforms traditional models in predicting periodontitis stage and grade.
• This NLP-driven approach demonstrates a groundbreaking application in dentistry for CDSS, enhancing diagnostic capabilities.
• The integration of advanced NLP with CDSS promises timely interventions, reduced healthcare costs, and improved patient outcomes.