"Retrospective Study of the Risk of Pathology in the Mandibular Second Molar Associated With an Impacted Third Molar: Development and Validation of a Semi-Automated Artificial Intelligence-Based Digitisation Platform". (SMART-M3)

Design. Single-centre, observational, cross-sectional, retrospective study approved by the Ethics Committee of the Universidad Católica San Antonio de Murcia (code CE102517). It follows the Declaration of Helsinki. Informed consent was waived because only pre-existing images obtained for clinical purposes are used. Data are pseudonymised in accordance with Regulation (EU) 2016/679 (GDPR) and Spanish Organic Law 3/2018.

Setting and sample. Consecutive sampling of all eligible images acquired at the UCAM Dental University Clinic between February 2018 and June 2026 with an ORTHOPHOS SL 3D unit (Dentsply Sirona). The unit of analysis is the mandibular third molar, so each patient may contribute one to four records (one or two third molars, on OPG, CBCT or both). The expected sample is approximately 2,110 third molars from 1,188 patients on OPG and approximately 690 third molars from 384 patients on CBCT, with a paired OPG-CBCT subsample of 667 teeth.

Variables. Third molar variables are:

• Winter classification
• Pell & Gregory depth and ramus relationship
• Retromolar space
• Angle between the long axes of the second and third molars
• Distance from the distal surface of the second molar to the ramus
• Mesiodistal diameter of the third molar
• Presence, type and level of contact between the third and second molars Covariates are age, sex and smoking status. Outcome. The primary outcome is a composite of distal pathology of the mandibular second molar, defined as the presence of at least one of the following: distal caries, distal bone loss or external root resorption. Teeth without any of the three are classified as controls. Each component is also recorded separately, with its extent or severity.

Statistical analysis. The analysis comprises:

• Descriptive statistics, overall and by imaging modality.
• Bivariate analysis using chi-square or Fisher's exact tests and Mann-Whitney U or Student's t tests, with odds ratios, 95% confidence intervals and correction for multiple comparisons.
• Multivariable mixed-effects logistic regression with the patient as a random effect, to account for the clustering of teeth within patients.
• Assessment of the model's discrimination with ROC curves and AUC.
• Agreement between OPG and CBCT in the paired subsample, using intraclass correlation coefficients, simple and weighted kappa and Bland-Altman analysis, with 95% confidence intervals obtained by patient-clustered bootstrap.
• Intra-observer reliability on 123 teeth per modality. AI-based platform. A semi-automated digitisation platform based on artificial intelligence (automatic landmark detection) is developed to perform the radiographic measurements. It is validated against expert manual assessment in an independent subsample of 191 third molars (103 OPG and 88 CBCT).