Comparative Evaluation Of Conventional Clinical Methods And Artificial Intelligence-Based Diagnostic Models In The Detection Of Chronic Generalized Gingivitis

Gingivitis is one of the most prevalent periodontal diseases worldwide, and early, accurate diagnosis is critical for preventing progression to irreversible periodontitis. While conventional diagnostic indices such as the Gingival Index (GI) and Plaque Index (PI) are widely used, they are inherently subjective and examiner-dependent. Recent advances in artificial intelligence (AI) offer promising alternatives for objective, scalable, and efficient periodontal diagnostics. Aim:To compare the diagnostic accuracy, sensitivity, and consistency of conventional clinical methods with an Artificial Neural Network (ANN)-based AI model in evaluating gingival health across a large population. Materials and Methods: An in vivo comparative study was conducted on 1,000 patients aged 14 to 75 years at a tertiary dental care institution. Gingival condition was evaluated using the GI and PI, and scored for color, contour, interdental papilla form, and presence of calculus. Standardized intraoral images were captured and analyzed using a trained ANN model comprising four specialized sub-networks. Diagnostic scores from AI and clinical methods were statistically compared using t-tests, ANOVA, correlation matrices, and reliability testing. Results:Conventional methods diagnosed gingivitis in 100% of patients with perfect inter-parameter consistency (Cronbach’s α = 1.0). The ANN model demonstrated high sensitivity (99.8%) and good internal reliability (Cronbach’s α = 0.784), particularly in detecting interdental papilla (accuracy: 74.91%) and calculus (accuracy: 80.47%). However, the AI system underperformed in identifying gingival color changes (accuracy: 70.78%; consistency: 63.42%) and showed weaker correlation in its provisional classification (r = 0.044). Significant statistical differences (p < 0.001) were observed across all diagnostic categories between AI and conventional methods. Conclusion:AI-assisted gingivitis detection demonstrates high sensitivity and consistency, especially for morphological parameters, and holds strong potential as an adjunctive tool in periodontal diagnosis. However, limitations in detecting early color changes and calculus suggest the need for algorithmic refinement and dataset expansion. AI models, when optimized and integrated with clinical workflows, may serve as scalable, objective complements to traditional diagnostic methods in both routine and community-based periodontal care.