An observational study in Peri-implantitis, Alveolar Bone Loss and Dental Implant, sponsored by Ankara University. Not yet recruiting at 1 site in Turkey (Türkiye). Open to participants aged 18 Years and older, including healthy volunteers. Per ClinicalTrials.gov, last updated 2026-10-30.
Sponsored by Ankara University · Observational
Dental implants are a routine treatment for missing teeth, but the number of implants affected by peri-implantitis, a progressive inflammatory disease of the tissues around implants, is increasing. The main radiographic sign of peri-implantitis is marginal bone loss around the implant. Early detection and correct grading of this bone loss in proportion to implant length are important for treatment planning and prognosis.
Panoramic radiographs are widely used for general screening in dental practice, but judging bone loss on them is subjective, time-consuming and depends on the clinician's experience. Multimodal large language models (vision-language models) are increasingly able to interpret images, yet their ability to evaluate peri-implant bone loss on panoramic radiographs has not been tested.
This single-center, retrospective, observational diagnostic accuracy study will evaluate how accurately five publicly available large language models (ChatGPT, Gemini, Claude, DeepSeek and Grok) detect and grade peri-implant marginal bone loss on anonymized, cropped panoramic radiographs of functionally loaded dental implants, compared with a reference standard established by three blinded clinicians (two periodontologists and one oral and maxillofacial radiologist). Bone loss will be graded with the Froum and Rosen (2012) classification: Score 0 (no visible bone loss), Score 1 (\<25% of implant length), Score 2 (25-50%) and Score 3 (>50%).
Approximately 784 implant sites (392 with and 392 without marginal bone loss) from existing panoramic radiographs in the hospital archive will be included. No new radiographs will be taken, no patient will be contacted and no treatment will be given; informed consent has been waived by the ethics committee. All images will be fully anonymized before evaluation.
DESIGN AND SETTING Single-center, retrospective, observational diagnostic accuracy study conducted at Ankara University Faculty of Dentistry (Departments of Periodontology and Oral and Maxillofacial Radiology), Ankara, Türkiye. The study is reported according to STARD 2015.
IMAGE ACQUISITION AND ANONYMIZATION Panoramic radiographs taken during routine diagnosis and treatment will be retrieved retrospectively from the hospital information system/PACS archive. Before any evaluation, the data curator will remove all patient identifiers (name, national ID number, protocol number, acquisition date). Images will be exported in PNG format and each implant with its surrounding bone will be cropped to a standardized region of interest (ROI) using ImageJ. Each radiograph and each ROI will receive a random sequential code (e.g., R-001, IMP-001). Only anonymized, coded ROI images will be used in all subsequent steps.
REFERENCE STANDARD Three blinded investigators will independently grade each ROI according to the Froum and Rosen (2012) peri-implantitis classification, which expresses marginal bone loss as a proportion of implant length: Score 0, no visible marginal bone loss; Score 1 (mild), bone loss \<25% of implant length; Score 2 (moderate), 25-50%; Score 3 (severe), >50%. A proportional index was chosen instead of millimetric measurement because it is robust to magnification differences on panoramic radiographs and accounts for implant length. Inter-rater reliability will be assessed with Fleiss' kappa. Disagreements will be resolved by consensus; the consensus score will serve as the reference standard ("ground truth").
INDEX TESTS Anonymized ROI images will be uploaded to the current secure/enterprise interfaces of five multimodal large language models (ChatGPT, Gemini, Claude, DeepSeek, Grok) with settings that prevent use of the images for model training. To avoid prompt-engineering bias, all models will receive an identical standardized zero-shot prompt (in Turkish) that (1) assigns the role of an expert periodontologist and dentomaxillofacial radiologist, (2) asks the model to examine the implant and marginal bone level in the attached panoramic radiograph section, (3) instructs it to grade bone loss as a proportion of total implant length, and (4) requires the output to be only one of Score 0, 1, 2 or 3 with no additional explanation. Model outputs will be recorded directly without modification.
SAMPLE SIZE An a priori power analysis (G*Power 3.1; exact binomial test, two-tailed, effect size g = 0.071, alpha = 0.05, power = 0.80, constant proportion = 0.50) based on Liu et al. (BMC Oral Health 2022;22:11) yielded a minimum of 392 implant sites per group; a balanced dataset of at least 784 implant sites (392 with and 392 without marginal bone loss) will be collected. Because one radiograph may contain several implants, the number of patients/radiographs will be smaller than the number of implant sites.
STATISTICAL ANALYSIS Sensitivity, specificity, positive and negative predictive values and overall accuracy of each model will be calculated against the reference standard, with 95% confidence intervals. Agreement between model outputs and the reference standard will be assessed with kappa statistics, differences between model and reference classifications with McNemar's test, and discriminative performance with ROC analysis. Analyses will be performed in SPSS and Python.
ETHICS Approved by the Ankara University Faculty of Dentistry Ethics Committee for Non-Clinical Scientific Research (decision no. 25/12). Informed consent was waived because the study is a retrospective archive review with no patient contact or intervention. No external funding.
315 studies on the registry are indexed under Peri-Implantitis; 87 are open to participants now.
This study's planned enrollment of 784 is above the median of 80 across 117 observational studies indexed under Peri-Implantitis.
Browse Peri-Implantitis studies →Ankara University is the lead sponsor of 320 studies on the registry; 64 are open to participants now.
Counted across the registry records on this site, refreshed daily.
Anonymized panoramic radiographs retrieved retrospectively from the PACS archive of Ankara University Faculty of Dentistry, taken for diagnostic purposes in adult patients (18 years or older) of either sex who have at least one functionally loaded dental implant. The unit of analysis is the implant site (standardized cropped region of interest); one radiograph may contribute more than one implant site. Consecutive eligible radiographs will be screened until at least 392 implant sites with and 392 implant sites without marginal bone loss are obtained.
Go through the criteria one by one — yes, no or not sure — and take the list to your doctor or the study team. Kept in this browser only.
Exclusion Criteria:
Anonymized, standardized region-of-interest (ROI) images of functionally loaded dental implants on panoramic radiographs showing peri-implant marginal bone loss (Froum and Rosen Score 1, 2 or 3) according to the consensus of three blinded clinicians. Target: 392 implant sites.
Diagnostic Test: Multimodal large language model evaluation (ChatGPT, Gemini, Claude, DeepSeek, Grok)
Anonymized, standardized ROI images of functionally loaded dental implants on panoramic radiographs with no visible peri-implant marginal bone loss (Froum and Rosen Score 0) according to the consensus of three blinded clinicians. Target: 392 implant sites.
Diagnostic Test: Multimodal large language model evaluation (ChatGPT, Gemini, Claude, DeepSeek, Grok)
Each anonymized ROI image is uploaded to five multimodal large language models using an identical standardized zero-shot prompt; the model assigns a Froum and Rosen score (0-3). Outputs are compared with the clinician reference standard.
Sensitivity and specificity of each large language model for detecting peri-implant marginal bone loss
For each of the five models (ChatGPT, Gemini, Claude, DeepSeek, Grok): sensitivity and specificity for the presence of marginal bone loss (Froum and Rosen Score 1-3 vs Score 0) on anonymized panoramic ROI images, calculated against the consensus reference standard of three blinded clinicians, with 95% confidence intervals.
Time frame: Single time point, at completion of model evaluation (approximately 8 weeks after study start); retrospective images, no follow-up
Positive predictive value, negative predictive value and overall accuracy of each model
PPV, NPV and overall accuracy of each of the five models for detecting marginal bone loss (present/absent), calculated against the reference standard with 95% confidence intervals.
Time frame: Single time point, at completion of model evaluation (approximately 8 weeks after study start)
Agreement between model severity grading and the reference standard
Agreement of the 4-level Froum and Rosen scores (0-3) assigned by each model with the consensus reference score, assessed with weighted kappa; disagreement between model and reference classification tested with McNemar's test.
Time frame: Single time point, at completion of model evaluation (approximately 8 weeks after study start)
Area under the ROC curve of each model
Discriminative performance of each model for marginal bone loss (present/absent), expressed as area under the receiver operating characteristic curve with 95% confidence interval.
Time frame: Single time point, at completion of model evaluation (approximately 8 weeks after study start)
Inter-rater reliability of the three blinded clinicians
Fleiss' kappa for the independent Froum and Rosen scores (0-3) assigned by the three clinicians before consensus.
Time frame: Single time point, at completion of clinician scoring (approximately 6 weeks after study start)
Plan to share: No — The ethics committee approval and study protocol state that data will be reported only in aggregate (statistical) form and will not be shared with third parties. The anonymized radiographic images and score data will therefore not be shared as individual participant data.
Start is estimated for Oct 2026; primary completion for Dec 2026, 2 months later.
How these figures are worked out, and which studies are left out of them.
From the registry record's own update history. This site started tracking changes on Sep 25, 2026; for anything earlier, see the record history on ClinicalTrials.gov ↗
This study is not yet recruiting, as verified in Oct 2026. You cannot join it, but the record below documents what was studied.
Get an email when the registry record changes — status, dates, results — or when someone posts here.
Sign in to followQuestions and observations about this study, from anyone following it. Not medical advice, and not a channel to the study team — their contact details are on the registry record.
Sign in to join the discussion. Reading takes no account; posting does. You choose a display name, and a pseudonym is the default.
Nothing here yet. If you are running this trial, taking part in it, or weighing whether to, this is the place to say so.