Aplicação de redes neurais na estimativa da idade de adultos a partir de imagens tomográficas de incisivos centrais superiores

Abstract

The estimation of chronological age using dental images is an essential practice in forensics dentistry, particularly for identifying individuals. This study employed convolutional neural networks (CNNs) to estimate the age of Brazilian adults based on tomographic images of maxillary central incisors. A total of 662 cone-beam computed tomography (CBCT) scans from patients aged 18 to 60 years were analyzed, amounting to 1,324 teeth. The process involved the training of four neural networks: the first (YOLOv8) identified and isolated the teeth in the original images; the second (YOLOv8) performed segmentation of the regions of interest (pulp chamber and dental crown); the third (ResNeXt-101) was trained to estimate age using regression based on coronal, sagittal, and concatenated tomographic slices (coronal and sagittal slices combined into a single image); and the fourth (ResNeXt-101) estimated age through regression based on the segmented images. The automated identification and segmentation of the regions of interest (pulp chamber and dental crown) achieved excellent performance metrics, with precision exceeding 0.9 for both the pulp chamber and dental crown. Regardless of the slice analyzed, the largest errors could be observed in older age groups. Additionally, the segmented images of specific areas of the pulp chamber and dental crown showed higher mean absolute errors (MAEs) compared to the detected tooth images. The lowest MAE, 4.99 years, was achieved through the analysis of concatenated detected images. The conclusion was that, although this study was limited to a specific population, the application of deep neural networks is an effective approach for estimating chronological age in forensic dentistry. This method overcomes the limitations of traditional manual methods, offering a more accurate and less bias-prone alternative.