AI Enabled Automated Digital Image Analysis for Peri‐Implant Health Monitoring: A Preliminary Diagnostic Study

OBJECTIVES: To develop and validate an AI-driven Digital Erythema Score (DES) for objective, non-invasive assessment of peri-implant mucosal erythema using intraoral scan (IOS) data. MATERIALS AND METHODS: An internal dataset (200 patients, 265 implants) was utilized, with the reference standard established via consensus by two independent experts. We developed a U-Net model to automatically segment the peri-implant mucosal margins to calculate the DES and diagnose erythema (modified Gingival Index mGI 0, ≥ 1). We used the YOLO model for object detection to localize erythema regions. The system's diagnostic performance was validated on an independent external dataset (23 patients, 30 implants), supplemented by a multi-observer study to evaluate the AI's assistance across clinicians with varying experience levels. RESULTS: The U-Net model achieved precise ROI segmentation (Dice similarity coefficient 0.69 ± 0.11) and reliable colorimetric calculation (ICC > 0.95). At the optimal threshold, the DES accurately identified mucosal erythema (AUC, internal 0.9228 [95% CI: 0.8533 to 0.9760], external 0.9018 [0.8135 to 0.9773]). DES and mGI showed significant correlations with bleeding on probing (B = 7.38, OR = 5.32, both p < 0.001). The YOLO model localized inflammatory regions (mAP internal 0.8968, external 0.6820). In the external validation, the DES-driven diagnostic framework enhanced the diagnostic performance of observers with different levels of experience and increased the inter-observer consistency. CONCLUSIONS: The DES provides an objective metric that converts chromatic signals into a diagnostic score and is significantly correlated with clinical inflammation. This AI-driven framework reduces clinical subjectivity and establishes a non-invasive paradigm for teledentistry and high-frequency, home-based monitoring of peri-implant health in the future after validation with patient-accessible imaging devices. TRIAL REGISTRATION: This study has been prospectively registered (NCT07349095).