ESTIMATING CHRONOLOGICAL AGE USING NEURAL NETWORKS: AN EVALUATION OF TOMOGRAPHIC IMAGES OF THE PULP CHAMBER

William José Lopes Junior; Alexandre Vieira Pereira Pacelli; Davi Magalhães Pereira; Heder Soares Bernardino; Marcelo Bernardes Vieira; Saulo Moraes Villela; Karina Lopes Devito

All Papers

Complete Paper

Summary

Paper Title

ESTIMATING CHRONOLOGICAL AGE USING NEURAL NETWORKS: AN EVALUATION OF TOMOGRAPHIC IMAGES OF THE PULP CHAMBER

Authors

William José Lopes Junior
Alexandre Vieira Pereira Pacelli
Davi Magalhães Pereira
Heder Soares Bernardino
Marcelo Bernardes Vieira
Saulo Moraes Villela
Karina Lopes Devito

Modality

Abstract for poster Presentation

Subject area

Mathematical modeling of biological phenomena associated to aging and/or cell senescence

Publishing Date

01/07/2024

Country of Publishing

Brasil

Language of Publishing

Inglês

Paper Page

https://www.even3.com.br/anais/barts-brazilian-aging-research-technology-symposium-2024/839336-estimating-chronological-age-using-neural-networks--an-evaluation-of-tomographic-images-of-the-pulp-chamber

ISBN

978-65-272-0505-0

Keywords

neural network, CNN, Age , forensic dentistry, tomography

Summary

Age estimation is an important demand in Forensic Dentistry, both to assist in the identification of living individuals (related to migratory movements, lack of documentation, memory loss, child support, underage marriage, legal responsibility, and cases of child pornography) and for the identification of corpses and human remains1. A widely known and accepted method for estimating chronological age is the evaluation of the pulp chamber dimension of permanent teeth2-3. The pulp chamber is a space inside the tooth crown, where the dental pulp is located, composed of nerves, blood vessels, and connective tissue. The entire chamber is surrounded by dentin, protected by enamel. As age advances, there is a reduction in the volume of the Pulp chamber due to the deposition of secondary dentin. This decrease can be observed in imaging exams and is an indicator of chronological age in adult individuals. Despite advances in radiographic exams and image evaluation software, the determination of pulp measurements and, consequently, age estimation, is still significantly influenced by the subjectivity of the image evaluator, as manual boundary marking and measurements may vary. In an attempt to reduce this interference, artificial intelligence models are being used to assist in age estimation4-7. Therefore, the present study aimed to develop convolutional neural networks (CNNs) to assist in estimating the age of adult Brazilians by evaluating the pulp area in cone beam computed tomography (CBCT) images of upper central incisor teeth. CBCT images of 554 healthy teeth, belonging to individuals from the Zona da Mata Mineira region (Minas Gerais, Brazil), of both sexes and aged between 18 and 60 years, were analyzed, divided into eight age groups. For evaluation, the most central coronal and sagittal sections of each tooth were used. These sections were subsequently cropped to reduce surrounding structures around the tooth. In addition to the original images, a “side by side” section was also created, allowing simultaneous evaluation of coronal and sagittal sections (Figure 1). To select the study's regions of interest, "tooth crown" and "pulp chamber" masks were manually created on coronal and sagittal sections of 194 teeth, which helped the neural networks to perform the segmentation process, using a proportion of 70% for network training, 15% for validation, and 15% for testing. For the age group classification network, images (coronal, sagittal, and “side by side”) of the remaining 360 teeth were used, using the same proportion as the segmentation network. Both for the segmentation process and classification, the YOLOv8-CNN8 was used, in ExtraLarge and Small variants, for segmentation and classification, respectively. The hyperparameter configuration was the same for both CNNs. The initial learning rate used is 1 × 10-3 , with a batch size of 16 and 160-pixel images. As for the optimizer, the "auto" parameter was used. The models were trained over 1000 epochs, without using pre-trained weights. To avoid overfitting, the Early Stopping in method was employed, with 150 epochs. During training, the following data augmentation techniques were implemented: modification of the HSV-Value component, translate, scale, and flip leftright. The development of CNNs was conducted through the Kaggle platform, based on the Python language. The following metrics were calculated to evaluate the performance of the created CNNs: accuracy, precision, recall, F-measure, and mean average precision (MAP). The average results obtained for the segmentation network were: precision of 0.983, recall of 0.998, and MAP of 0.994. For the classification networks, the following average (standard deviation) accuracies were obtained: 0.32 (0.05), 0.33 (0.04), and 0.34 (0.03), for coronal, sagittal, and “side by side” sections, respectively. CNNs can segment images very satisfactorily in their regions of interest, even with a reduced dataset. As for the classification network, although it demonstrates promising results, there is a need for a more robust dataset for the network to detect the subtle changes in images related to age variation.

Title of the Event: BARTS - II Brazilian Aging Research & Technology Symposium (2024)
City of the Event: Juiz de Fora
Title of the Proceedings of the event: Proceedings of the Second Brazilian Aging Research and Technology Symposium
Name of the Publisher: Even3
Means of Dissemination: Meio Digital

How to cite

JUNIOR, William José Lopes et al.. ESTIMATING CHRONOLOGICAL AGE USING NEURAL NETWORKS: AN EVALUATION OF TOMOGRAPHIC IMAGES OF THE PULP CHAMBER.. In: Proceedings of the Second Brazilian Aging Research and Technology Symposium. Anais...Juiz de Fora(MG) Universidade Federal de Juiz de Fora, 2024. Available in: https//www.even3.com.br/anais/barts-brazilian-aging-research-technology-symposium-2024/839336-ESTIMATING-CHRONOLOGICAL-AGE-USING-NEURAL-NETWORKS--AN-EVALUATION-OF-TOMOGRAPHIC-IMAGES-OF-THE-PULP-CHAMBER. Access in: 12/06/2025