•  
  •  
 

Abstract

Background/purpose: Conventional periodontal probing is the gold standard for evaluating periodontal disease severity; however, it is time-consuming and susceptible to operator-dependent variability. This study developed and clinically validated an artificial intelligence (AI)-assisted radiographic system for automated periodontal disease severity assessment using full-mouth radiographs, and investigated its association with clinical periodontal probing depth.

Materials and methods: A total of 655 full-mouth radiographs were retrospectively collected: 151 (3,962 teeth) for AI model training and 504 (13,809 teeth) for validation. Periodontal bone loss was quantified using the periodontal bone loss (PBL) ratio and classified as mild, moderate, or severe. Clinical validation used 31 radiographs with corresponding periodontal charting records (768 teeth). Pearson correlation, simple linear regression, one-way ANOVA, and receiver operating characteristic (ROC) analyses evaluated the relationship between AI-derived values and maximum probing depth (Max PD).

Results: The AI model achieved an overall classification accuracy of 90.48% (range: 90.5%–96.8% across severity categories). Misclassifications occurred only between adjacent severity levels. The severe category demonstrated the highest precision (96.5%) and specificity (99.0%); the moderate category achieved the highest sensitivity (98.5%). AI-derived values were moderately correlated with Max PD (r = 0.502, P < 0.001; R² = 0.252). ROC analysis yielded AUCs of 0.767 and 0.789 for Max PD ≥ 6 mm and ≥ 7 mm, respectively

Conclusion: The AI-assisted radiographic system demonstrated high diagnostic accuracy and significant correlation with clinical probing depth, supporting its role as an objective adjunctive tool for periodontal disease screening, severity stratification, and clinical decision support.

Publication Date

2026

Received Date

June 28 2026

Accepted Date

July 14 2026

Final Revision Date

July 14 2026

Download

Share

COinS