Sex determination and identification of population origin are two essential elements in forensic investigation, being fundamental tasks when dealing with human skeletal remains [14, 15]. Therefore, the aim of this study was to employ measurements of the dental arch and maxillary skeletal base to determine sex, using supervised machine learning. The construction of this model aims to enhance sex determination based on the maxillary complex, using features extracted from CBCT images. Although other craniofacial measurements have been used for sexual dimorphism assessment, such as the frontal bone, dental measurements, mandibular measurements, hyoid bone, and cervical vertebrae [4, 16,17,18,19,20], until now, no studies have been found in the literature employing maxillary measurements for sex determination using machine learning algorithms.

Machine learning involves the use of various mathematical models capable of generating predictive models through data analysis [20]. Developing a model to estimate sex is based on solving a classification task, which is a common task in machine learning [9]. One of the algorithms used for this task is Support Vector Machine (SVM), a kernel-based machine learning model employed for classification and regression tasks. The primary goal of the SVM algorithm is to find a hyperplane that maximizes the margin between different classes in the dataset [21]. SVM has proven effective in sex estimation based on cranial measurements, being considered one of the best-performing predictive models in forensic literature [8, 9, 22]. In this study, the SVM-based predictive model demonstrated the best performance during both the testing and cross-validation stages.

The presence of sexual dimorphism in the maxillary complex has been observed in several studies [6, 23, 24], demonstrating good predictive capacity in sex determination. Similarly, interdental transverse dimensions in the maxillary arch show dimorphic traits, in which it is larger in males than females [25, 26]. Among the evaluated measurements, skeletal dimensions showed good predictive power, while interdental dimensions also demonstrated a high effect size, with the intermolar distance exhibiting the largest effect among all evaluated dental variables. The NFW variable was the only one that showed a test power below 0.7. Although it is possible to hypothesize that a larger sample size could increase the observed test power, the vast majority of variables showed adequate power, thus validating the sample size used. Additionally, despite intermolar distance (Mol-Mol) showing the largest effect size, the Ln-Ln variable was the one that showed greater importance for most of the trained models, in which it was possible to estimate the importance of the variables. Extrapolating to the forensic context, skeletal characteristics can be considered more stable, since teeth may be subject to orthodontic movement, as well as prosthetic rehabilitations.

In general, females have lower bone mass compared to males, regardless of the age range assessed [27]. This sexual dimorphism becomes more evident due to changes in bone tissue caused by sex hormones (estrogens and androgens), genetics, and inflammatory processes, which affect the formation, resorption, and death of osteoclasts, making it a multifactorial process [4, 27]. Most of the bone and muscle mass is acquired before the age of 18. Sex hormones and the GH/IGF-1 axis regulate bone growth in both sexes, resulting in skeletal differences during puberty. As a result, men tend to have longer and wider bones due to differences in periosteal apposition and endosteal resorption, although bone mineralization is similar between the sexes [28]. This difference is also reflected in skull size [29]. In this regard, the univariate analysis of the assessed variables revealed statistical significance, demonstrating evident differences in maxillary measurements between men and women.

The representativeness of a sample in relation to a target population occurs when the estimates obtained, or their interpretation can be generalized to that target population. In the context of sexual dimorphism, it is important to consider factors such as the age range of individuals in the sample and the individual characteristics of each region. When examining the age of participants, statistically significant differences between sexes in the transverse widths of the maxillary skeleton were identified in age groups between 10 and 14 years, with male individuals generally showing greater incremental growth changes than female individuals. Additionally, the morphological characteristics of dental arches may vary according to ethnicity [6, 25]. In the present study, the age range of participants was from 10 to 88 years and was matched between groups in a 1:1 ratio to prevent age from acting as a confounding factor in the analysis. Although the inclusion of a broader age range may reduce the accuracy of the model, it also increases the external validity of the study, thus providing better applicability in the forensic context. In forensic settings, accurately determining age can be challenging due to the variability in the quality of available structures. While this decision might have decreased the precision of the models, it enhanced the external validity and applicability of our findings, making them more generalizable. Regarding ethnicity, it is important to mention that the data collection was carried out in the South region of Brazil, which is characterized by its unique miscegenation. Therefore, the sample consisted exclusively of Brazilians, which limits its representativeness. Therefore, it is recommended to conduct studies in other ethnic groups for a more comprehensive understanding of sexual dimorphism in these structures.

The bones of the head and neck are primarily originate from the branchial arches and develop around the fourth week of gestation [30]. As a result of this process, it is common for the structures of the male maxilla to be larger than those of the female maxilla in almost all dimensions, especially in length and total width of the maxilla [6, 23]. In the present study, measurements of the dental arch and maxillary skeletal base showed good predictive capacity for identifying sex. Additionally, the use of metric parameters was an approach adopted to minimize subjectivities, with all measurements performed in CBCT scans, which proved to be an excellent way to identify precise measurements, even considering the overlap of some bone structures. Although the variables used showed good predictive power, it is reasonable to suppose that even greater accuracy could be achieved with a larger sample size, and individuals from different ethnicities should be included in future studies. However, the results of the present study highlight the predictive potential of these variables, which are important markers for sexual dimorphism and may have relevant applicability in forensic science.