中国人における肘MRIとデータマイニングに基づく法医学的年齢推定および法的年齢閾値分類

This study developed and evaluated data mining models for age estimation and legal age thresholds classification based on elbow MRI in a contemporary Chinese population. A total of 867 patients (614 males and 253 females) aged 30 years or younger were retrospectively included, and T1-weighted coronal elbow MRI images were assessed by experts using five age-related indicators. Multiple data mining models were constructed, with mean absolute error (MAE) as the primary performance metric. Both intra- and inter-observer agreements demonstrated good reliability. In external validation, the optimal model achieved the lowest MAE of 2.9845 years in males and 2.9767 years in females, compared to 2.0471 years and 3.0477 years, respectively, in internal validation. For classification at the 12-, 14-, 16-, and 18-year threshold, the area under the receiver operating characteristic curve (AUC) exceeded 0.90 across all models. Notably, these models also attained the highest specificity (defined as the proportion of individuals truly below the threshold correctly identified) reaching 1.000, 1.000, 0.970, and 0.800 in males, and 1.000, 1.000, 0.857, and 0.750 in females at the respective thresholds. In conclusion, the proposed models show promise in estimating age and classifying legal thresholds, with performance that is generally comparable to that of other MRI-based approaches in certain contexts. These preliminary findings suggest that elbow MRI may serve as a useful tool for age assessment in the Chinese population, though further validation is warranted to confirm its generalizability.