Gender Classification Using Parameters Obtained from the Dens Axis with Machine Learning Algorithms and Multilayer Perceptron Classifier

Abstract

Background and Objectives: Due to the difficulties associated with the separation, damage, cremation, and commingling of skeletal remains, it is of great importance in forensic medicine to assess the accuracy and reliability of sex estimates derived from different skeletal components. For this purpose, this study aimed to classify gender using machine learning (ML) algorithms and a multilayer perceptron classifier (MLPC) based on morphometric data of the dens axis obtained from computed tomography (CT) images. Methods: Retrospectively, measurements were taken from CT images of 300 male and 300 female individuals aged between 18-65 years, including dens axis height (DAH), anteroposterior (APDDA) and anterosuperior lengths (ASDDA), dens axis angle (DAA), clivodental angle (CDA), and Boogard angle (BOO). Machine learning models such as Extra Tree Classifier (ETC), Random Forest (RF), Decision Tree (DT), Gaussian Naive Bayes (GaussianNB), k-Nearest Neighbors (k-NN), Linear Discriminant Analysis (LDA), Quadratic Discriminant Analysis (QDA), and Logistic Regression (LR) were used. MLPC was chosen as artificial neural networks (ANN) model. Results: Significant differences were found between genders in all dens axis parameters except BOO (p<0.05). The highest accuracy rate in ML algorithm modeling was found to be 0.80 with LDA, RF, k-NN algorithms, and MLPC. The parameter with the highest impact on gender classification was the dens axis anterosuperior length. Conclusion: It was found that the parameters obtained from the dens axis using MLCP and ML algorithms have sufficient accuracy rates the classification of sex. It was concluded that in forensic medicine, in cases of deterioration, loss, and deficiencies in bone sources for biological identity determination, the morphometric features of the dens axis can be considered for gender prediction.