Explainable Deep Learning Approaches for Risk Screening of Periodontitis

Area of Science:

• Computational biology and bioinformatics
• Public health and epidemiology
• Artificial intelligence in healthcare

Background:

• Periodontitis is linked to systemic diseases, yet early screening tools are lacking.
• Effective prevention and early diagnosis are crucial for managing periodontitis.
• Existing methods for periodontitis screening require improvement for wider application.

Purpose of the Study:

• To develop and validate an explainable artificial intelligence (XAI) tool for early periodontitis screening.
• To identify key clinical, demographic, and biochemical factors associated with periodontitis risk.
• To provide individualized risk assessments for periodontitis using AI-driven insights.

Main Methods:

• Utilized a large dataset (30,465 participants) from the National Health and Nutrition Examination Survey (NHANES).
• Trained deep learning and machine learning models for periodontitis screening and diagnosis.
• Applied Local Interpretable Model-Agnostic Explanations (LIME) to identify and rank associated risk factors.

Main Results:

• Deep learning models achieved high performance (AUC 0.858-0.865) in screening and diagnosis.
• XAI identified crucial associated factors including age, sex, diabetes, tissue transglutaminase, and smoking.
• Other contributing factors identified include arthritis, sleep disorders, hypertension, cholesterol, and overweight status.

Conclusions:

• XAI effectively identifies key factors associated with periodontitis, aligning with clinical knowledge.
• The developed AI tool demonstrates utility in analyzing associated factors for periodontitis.
• XAI can support the development of early detection and prevention strategies for periodontitis during medical checkups.