Radiographic legal age estimation based on third molar development using Machine-learning algorithms

BACKGROUND: Age assessment plays an important role in forensic sciences to aid in criminal or civil matters. Third molars continue developing during the legal age of adulthood. Dental age estimation based on radiographic examination is an operator-dependent procedure. The aim was to test the performance of supervised machine learning (SML) to classify individuals according to the threshold of 18 years using the third molar index (I3m). METHODS: Panoramic radiographs (n = 597) of 52% males and 48% females between 13 and 26 years were used. Three Convolutional Neural Networks were compared to segment the third molar. Several machine learning algorithms were tested with a 10-fold cross-validation approach to identify the best age classification algorithm. The SML age estimation model was compared to the expert. RESULTS: Attention U-Net achieved the highest segmentation scores, and the k-nearest neighbors (KNN) showed the best scores of classification algorithms. The SML (KNN) scored sensitivity = 77.77%, specificity = 96%, and AUC = 0.87 in males. In females, sensitivity = 77.4%, specificity = 86.7%, and AUC = 0.82. The expert scores were: sensitivity = 70.4%, specificity = 100%, AUC = 0.85 for male, and for female: sensitivity = 54.8%, specificity = 93.3%, AUC = 0.74. CONCLUSION: SML classified males and females below and above the legal age with high accuracy compared to manual methods. The presented work suggests that AI, using pixel-level analysis, can detect subtle cues beyond human perception. This signals a possible departure from traditional age estimation indices.