Minimum age and diagnostic accuracy approaches in knee MRI-based forensic age estimation: A comparative and combined...

This study aimed to evaluate the performance of the Schmeling-Kellinghaus and Dedouit staging systems for forensic age estimation based on magnetic resonance imaging (MRI) of the distal femoral and proximal tibial epiphyses, within both a minimum age framework and a diagnostic-accuracy (ROC-based binary classification) framework. In this retrospective study, MRI examinations from a single French university hospital acquired between 2024 and 2025 were analyzed. For both staging systems, intra- and interobserver agreement demonstrated very good reliability. Strong, statistically significant positive Spearman correlations between ossification stage and chronological age were observed for the overall sample and for each sex separately (overall rho 0.84-0.88, all p < 0.001). Sex-related differences in ossification timing, assessed with the Mann-Whitney U test and a Holm correction for multiple comparisons, were confirmed at intermediate-to-advanced stages of both systems, with females reaching each stage at systematically younger ages than males. Diagnostic performance was evaluated using repeated (5 × 10-fold) cross-validated logistic regression, so that both threshold selection (via the Youden index) and performance evaluation were performed out-of-sample; 95% confidence intervals for AUC (DeLong method) and for sensitivity, specificity, and accuracy (bootstrap) are reported alongside the number of positive and negative cases at each threshold. At the 18-year threshold, cross-validated AUC reached 0.921 (Schmeling-Kellinghaus) and 0.960 (combined model) in the femur, and 0.812 and 0.922, respectively, in the tibia. Across both anatomical regions and all three thresholds, the combined model achieved significantly higher AUC than either single system in 11 of 12 DeLong comparisons (p < 0.05); the one exception was Dedouit vs combined at the tibial 21-year threshold (p = 0.051). At the 21-year threshold, both single systems showed a marked ceiling effect: because complete epiphyseal fusion (Schmeling-Kellinghaus stage 4) is frequently attained well before 21 years, sensitivity for this threshold reached 1.000 in both regions, but at the cost of substantially reduced specificity (0.753 in the femur, 0.552 in the tibia); the combined model materially improved specificity at the tibial 21-year threshold (0.851) while retaining high sensitivity. In conclusion, both the Schmeling-Kellinghaus and Dedouit staging systems provide reliable and reproducible lower-bound age estimates in this cohort, and their statistical combination measurably improves diagnostic accuracy, particularly at intermediate legal thresholds and in the proximal tibia. Cross-validated, interval-reported diagnostic accuracy metrics could provide a more conservative and legally defensible framework for medico-legal decision-making than apparent (in-sample) performance estimates.