K05
Conditions
Interventions
Group 1: Single-arm complete enumeration of all patients of the department for whom a statutory periodontal treatment plan (BEMA) was submitted and approved by the health insurer since 1 July 2021 (n
Sponsors
LMU Klinikum
Eligibility
Sex/Gender
All
Inclusion criteria
Inclusion criteria: Patients of the Department of Conservative Dentistry, Periodontology and Digital Dentistry, LMU Munich. Statutory periodontal treatment plan (BEMA) submitted and approved by the responsible health insurer since 1 July 2021. Complete documented six-point clinical periodontal chart including diagnosis according to the 2018 classification (staging and grading). Availability of a digital panoramic radiograph obtained no more than 12 months before the date of clinical charting. Adequate diagnostic image quality of the panoramic radiograph. All criteria refer exclusively to pre-existing data collected during medically indicated diagnostics and treatment. The analysis is performed at tooth level; implant sites are excluded from analysis.
Exclusion criteria
Exclusion criteria: Missing or incompletely documented six-point clinical periodontal chart. No digital panoramic radiograph available, or an interval of more than 12 months between the radiograph and clinical charting. Panoramic radiograph of insufficient diagnostic image quality, for example due to positioning or motion artefacts, over- or underexposure, or incomplete depiction of the jaws. Documented objection to the further use of treatment data for research purposes. Edentulous jaw or no evaluable tooth remaining after application of the tooth-level criteria.
Design outcomes
Primary
| Measure | Time frame |
|---|---|
| Diagnostic performance of the deep learning model in categorically estimating the maximum mesial and maximum distal periodontal probing pocket depth per tooth (shallow 3 mm or less, moderate 4 to 5 mm, deep 6 mm or more) from the panoramic radiograph, measured as the area under the receiver operating characteristic curve (AUROC), reported class-wise (one-vs-rest) and as macro-average. The reference standard is the documented six-point clinical periodontal chart. Performance is reported exclusively on the previously unused, patient-stratified hold-out test set. Confidence intervals are obtained by bootstrapping. Time of assessment: single analysis of retrospectively available data from 1 January 2021 to 1 May 2026; there is no follow-up. Radiograph and clinical reference examination are separated by no more than 12 months. | — |
Secondary
| Measure | Time frame |
|---|---|
| Additional performance metrics according to the DentalCOMS recommendations for computer vision studies in dentistry: area under the precision-recall curve (AUPRC), precision (positive predictive value), sensitivity, specificity and F1 score, together with full confusion matrices, each reported class-wise (one-vs-rest) and as macro- and micro-average, with bootstrap confidence intervals. Diagnostic performance stratified by tooth type (anterior versus posterior teeth). Diagnostic performance stratified by jaw (maxilla versus mandible). Diagnostic performance stratified by site (mesial versus distal). Qualitative and quantitative evaluation of Grad-CAM activation maps to assess whether model decisions correspond to anatomically plausible regions (marginal bone level, interproximal bone defect). Time of assessment for all secondary outcomes: identical to the primary outcome, single analysis without follow-up. | — |
Countries
Germany
Contacts
Public ContactMatthias Folwaczny
Klinikum der Universität München (LMU Klinikum), Anstalt des öffentlichen Rechts Poliklinik für Zahnerhaltung, Parodontologie und Digitale Zahnmedizin
Outcome results
None listed