AI-Based Prediction of Root Coverage Outcome From Intraoral Photographs

Development and Internal Validation of a Deep Learning Model Predicting the Outcome of Root Coverage Surgery From Preoperative Intraoral Photographs: A Prospective Observational Cohort Study

This study evaluates whether the outcome of root coverage surgery can be predicted from a preoperative intraoral photograph. Adults with Cairo RT1,RT2 or RT3 gingival recessions treated with a coronally advanced flap and a connective tissue graft are followed for six months. Standardised photographs and clinical measurements are obtained before surgery and at each follow-up visit. A deep learning model is developed to predict the surgical outcome from the preoperative photograph and baseline clinical variables, and its performance is compared with the outcome measured clinically at six months. The model does not influence treatment decisions.

Study Overview

• Status: Recruiting
• Conditions: Gingival Recessions
• Intervention / Treatment: Procedure: Coronally advanced flap with subepithelial connective tissue graft; Diagnostic test: Deep learning based prediction of root coverage outcome

Detailed Description

Whether an exposed root surface can be completely covered is the central question in planning mucogingival surgery. The Cairo classification is the current diagnostic standard for that judgement, but assignment of the recession type varies between examiners and prediction of the individual surgical outcome remains largely subjective. In this cohort, consecutive systemically healthy adults with Cairo RT1,RT2 or RT3 gingival recessions are treated by a single operator with a coronally advanced flap combined with a subepithelial connective tissue graft. Recession depth, keratinised tissue width and gingival thickness are recorded at baseline and at three and six months.

Standardised intraoral photographs are obtained at each time point under fixed conditions. A deep learning model is developed to predict the six-month outcome from the preoperative photograph together with baseline clinical variables. Model performance is assessed by discrimination, calibration and prediction error, using the clinical measurement at six months as the reference standard. A secondary analysis examines whether the recession type assigned automatically from the photograph agrees with the type assigned by the examining periodontist. The model is developed and validated internally within this cohort; no external validation set is available. Its output is not shown to the operator and does not influence treatment. Reporting follows the TRIPOD recommendations for prediction model studies.

• Study Type: Observational
• Enrollment (Estimated): 36

Contacts and Locations

• Study Contact: Name: Muhammed F Dogan, Resident; Phone Number: +905433890065; Email: mfurkandogaan@gmail.com

Study Locations

• Turkey (Türkiye)
  • Istanbul
    • Istanbul, Istanbul, Turkey (Türkiye), 34854
      • Recruiting
      • Marmara University Faculty of Dentistry Department of Periodontology
      • Contact: Muhammed F Dogan, Resident; Phone Number: +905433890065; Email: mfurkandogaan@gmail.com

Participation Criteria

Eligibility Criteria

• Ages Eligible for Study: Adult; Older Adult
• Accepts Healthy Volunteers: Yes

• Sampling Method: Non-Probability Sample

Study Population

The study population consists of systemically healthy adult patients (aged 18-65) presenting to the Department of Periodontology at Marmara University with esthetic concerns or dentin hypersensitivity associated with gingival recession. The cohort includes individuals diagnosed with Cairo Class RT1 or RT2 (Miller Class I or II) gingival recession defects who are scheduled to undergo mucogingival root coverage surgery.

Description

Inclusion Criteria:

• Systemically healthy patients (ASA I or II status) with no contraindications for periodontal surgery.
• Adult patients aged 18 to 65 years.
• Presence of isolated or multiple gingival recessions classified as Cairo RT1, RT2 or RT3 in the maxilla or mandible.
• Patients with good oral hygiene standards, defined as a Full Mouth Plaque Score (FMPS) and Full Mouth Bleeding Score (FMBS) of < 20% at baseline.
• Presence of an identifiable Cemento-Enamel Junction (CEJ) (Crucial for AI segmentation).

Exclusion Criteria:

• Patients with uncontrolled diabetes, immune system disorders, or pregnant/lactating women.
• Teeth with cervical restorations or abrasions that obscure the CEJ.
• Malpositioned or rotated teeth that would distort the photographic angle for AI analysis.

Study Plan

How is the study designed?

Number of groups / cohorts

1

Cohorts and Interventions

Group / Cohort

Intervention / Treatment

Root coverage surgery cohort; Systemically healthy adults aged 18 to 65 years with Cairo RT1,RT2 pr RT3 gingival recessions, treated with a coronally advanced flap combined with a subepithelial connective tissue graft by a single operator and followed for six months. All participants received the same surgical technique; no comparison group was formed and no participant was assigned to a treatment for the purposes of this study. Standardised intraoral photographs and clinical measurements were obtained before surgery and at three and six months.

Procedure: Coronally advanced flap with subepithelial connective tissue graft; A coronally advanced flap is raised over the recession defect and a subepithelial connective tissue graft harvested from the palate is positioned beneath it, after which the flap is sutured coronal to the cemento-enamel junction. Graft thickness, length and width are recorded for each treated site. The procedure was performed as routine clinical care and was not assigned for research purposes.; Diagnostic test: Deep learning based prediction of root coverage outcome; Preoperative intraoral photographs and baseline clinical variables are analysed by a deep learning model that predicts the outcome of root coverage surgery. The model output is not used in clinical decision making and does not influence treatment; it is compared retrospectively with the outcome measured by the treating periodontist at six months. The same photographs are also used to assign the recession type automatically, which is compared with the clinical assignment.

What is the study measuring?

Primary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Accuracy of the model in predicting root coverage at six months	Difference between the root coverage predicted by a model based on preoperative intraoral photographs and baseline clinical characteristics, and the root coverage observed at six months. Root coverage is expressed as the percentage of the baseline recession depth that is covered, calculated as [(baseline recession depth - six-month recession depth) / baseline recession depth] × 100, from probing measurements made by the treating periodontist from the cemento-enamel junction to the gingival margin. Predictive accuracy is summarised as the mean absolute error in percentage points across all treated sites.	6 months

Secondary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Sensitivity and specificity of the model at the selected decision threshold	Proportion of sites correctly identified by the model among those that achieved the outcome (sensitivity) and among those that did not (specificity), evaluated at the operating point selected on the receiver operating characteristic curve. Both proportions are reported with 95% confidence intervals. The reference standard is the clinical measurement made at six months by the treating periodontist, using a periodontal probe from the cemento-enamel junction to the gingival margin.	6 months

Other Outcome Measures

Outcome Measure	Measure Description	Time Frame
Calibration of the model	Agreement between the probability predicted by the model and the frequency observed in the cohort, assessed by the calibration slope and intercept and displayed as a calibration plot. Discrimination indicates whether the model ranks sites correctly; calibration indicates whether the predicted probabilities are numerically correct, and the two are reported separately because a model may rank well while producing miscalibrated probabilities. The reference standard is the clinical measurement made at six months.	6 months
Agreement between the model-assigned and the clinician-assigned recession type	Proportion of treated sites at which the recession type assigned by the model from the preoperative photograph matches the type assigned by the examining periodontist, reported together with quadratic weighted kappa. The reference standard is the clinical assignment recorded at baseline according to the criteria of Cairo et al.	Baseline

Collaborators and Investigators

• Sponsor: Marmara University
• Investigators: Principal Investigator: Leyla Kuru, Professor, Marmara University Faculty of Dentistry Department of Periodontology

Study record dates

Study Major Dates

• Study Start (Actual): September 17, 2025
• Primary Completion (Estimated): September 17, 2026
• Study Completion (Estimated): September 17, 2027

Study Registration Dates

• First Submitted: August 16, 2026
• First Submitted That Met QC Criteria: August 16, 2026
• First Posted (Actual): August 20, 2026

Study Record Updates

• Last Update Posted (Actual): August 20, 2026
• Last Update Submitted That Met QC Criteria: August 16, 2026
• Last Verified: August 1, 2026

More Information

Terms related to this study

• Keywords: Artificial Intelligence; Deep Learning; Gingival Recession; prognostic model; Image Analysis
• Additional Relevant MeSH Terms: Periodontal Atrophy; Periodontal Diseases; Mouth Diseases; Stomatognathic Diseases; Gingival Diseases; Gingival Recession
• Other Study ID Numbers: 2025-09-02/2025-58

Drug and device information, study documents

• Studies a U.S. FDA-regulated drug product: No
• Studies a U.S. FDA-regulated device product: No