Oliveira Julyana de Araújo, Borella Natália Rogério, Ramos-Perez Flávia Maria de Moraes, Pontual Andrea Dos Anjos, Calazans Maria Alice Andrade, Ferreira Felipe Alberto Barbosa Simão, Madeiro Francisco, Pontual Maria Luiza Dos Anjos
Universidade Federal de Pernambuco - UFPE, Programa de Pós-Graduação em Odontologia, Recife, PE, Brasil.
Universidade Federal de Pernambuco - UFPE, Programa de Pós-Graduação em Engenharia Elétrica, Recife, PE, Brasil.
J Appl Oral Sci. 2025 Jun 27;33:e20250049. doi: 10.1590/1678-7757-2025-0049. eCollection 2025.
this study aims to evaluate the sexual dimorphism of the morphometric features of the frontal sinus via cone beam computed tomography (CBCT) reconstructions, using a predictive formula and an artificial neural network (ANN).
the morphometric features of the frontal sinuses obtained from 1,000 CBCT scans, equally distributed by sex, were assessed by two examiners. The frontal sinus morphometric features from 800 CBCT scans were analyzed using Mann-Whitney tests and a multivariate logistic regression model to identify key morphometric features for sex determination and to develop the predictive formula. These features were subsequently used to validate the predictive formula and the machine learning-based classification system. The predictive formula was evaluated using a set of 200 CBCT scans. The machine learning-based classification system consisted of a three-layer ANN trained with 80% of the CBCT scans and tested with the remaining 20%.
Except for the higher frontal sinus index in females, males exhibited higher numerical values for height, width, and anteroposterior (AP) length. The significance level for all statistical tests was set at 0.05. Multivariate logistic regression identified the following four essential morphometric features: sinus height, anteroposterior length (depth) of the sinus, sinus width, and total sinus width. Both the predictive formula and the ANN demonstrated sexual dimorphism. The accuracy, specificity, sensitivity, precision, and F1- score values were 73.50%, 74.00%, 73.00%, 73.74%, and 73.37% for the regression model, and 76.00%, 84.00%, 68.00%, 80.95%, and 73.91% for the ANN, respectively. Except for sensitivity, the ANN outperformed the predictive formula regarding maximum specificity, accuracy, precision, and F1 score.
both methods, particularly the ANN, can potentially support sex estimation in the Brazilian forensic context.
本研究旨在通过锥形束计算机断层扫描（CBCT）重建，利用预测公式和人工神经网络（ANN）评估额窦形态特征的性别差异。
由两名检查人员评估从1000例CBCT扫描中获得的额窦形态特征，这些扫描按性别平均分布。使用曼-惠特尼检验和多变量逻辑回归模型分析800例CBCT扫描的额窦形态特征，以确定用于性别判定的关键形态特征并建立预测公式。随后使用这些特征来验证预测公式和基于机器学习的分类系统。使用一组200例CBCT扫描评估预测公式。基于机器学习的分类系统由一个三层人工神经网络组成，该网络用80%的CBCT扫描进行训练，并用其余20%进行测试。
除女性的额窦指数较高外，男性在高度、宽度和前后（AP）长度方面的数值较高。所有统计检验的显著性水平设定为0.05。多变量逻辑回归确定了以下四个基本形态特征：窦高度、窦的前后长度（深度）、窦宽度和窦总宽度。预测公式和人工神经网络均显示出性别差异。回归模型的准确性、特异性、敏感性、精确性和F1分数值分别为73.50%、74.00%、73.00%、73.74%和73.37%，人工神经网络的相应值分别为76.00%、84.00%、68.00%、80.95%和73.91%。除敏感性外，在最大特异性、准确性、精确性和F1分数方面，人工神经网络的表现优于预测公式。
这两种方法，尤其是人工神经网络，在巴西法医背景下可能有助于性别估计。
