Sex Estimation through the Study of Computed Tomography of the Mandible

Abstract

Objectives

This study investigated the potential of the mandible as an anatomical structure for sex estimation using three-dimensional images generated by computed tomography.

Materials and Methods

The research was conducted on 93 dry mandibles from the ICB-USP collection, aged over 18 years, and analyzed through 12 external measurements performed in the Meshmixer® software.

Results

The data were subjected to binomial logistic regression and ROC curve analysis, revealing areas under the curve between 0.61 and 0.76, sensitivity between 55.36% and 80.36%, and specificity between 51.35% and 78.38% with ramus length, mandibular body height, and angle width as the most accurate. The study reinforces the feasibility of using tomography as an alternative to direct physical analysis of bones, especially when access to osteological specimens is limited. It highlights the importance of validating methods in multiracial populations such as the Brazilian.

Conclusion

It is concluded that mandibular anthropometric analysis by tomographic images is a complementary tool in forensic sex identification, with practical applicability when recovery of complete skeletal material is difficult.

Introduction

The process of identifying an individual postmortem involves several steps. In many cases, primary identification methods (DNA analysis, fingerprinting, and forensic dentistry) are unfeasible due to decomposition, the conditions of the place where the body was found, and in mass disasters, homicides, or partial bone findings (1).

Forensic anthropology plays a well-established role in international studies in basic partial identifications, such as biological sex, geographic affinity, height, and age (2-4). In cases of advanced decomposition, the mandible may detach from the skull, making identification difficult by methods that use anthroposcopic and anthropometric analyses of the skull and mandible together (5).

Sex estimation is one of the initial steps in identifying bone findings and is essential for the final identification of an individual (6). Several studies use anthropometric analyses to estimate sex, mainly applicable to the pelvis, which is considered the most efficient bone region for such estimation in forensic anthropology (7–10). However, not all bones are always found. Skull bones are also widely studied, and several methods have been developed and validated in different populations to expand the techniques applicable to different identification scenarios (11-13).

The mandible is a unique bone with several points of sexual dimorphism (14). Qualitative analyses can be subjective and at risk of interobserver error due to the difficulty in establishing consistent parameters (4, 7). Quantitative methods have been applied to samples from different populations, with a greater chance of achieving correct identification (15–22). Access to collections containing mandibles worldwide is limited, making them valuable for studies using anthropological methods. The diversity of studies and the application of different methods are essential in this context (23).

With the advancement of technology and greater access to it, computed tomography (CT) has been used as an auxiliary and facilitating method to estimate sex (24, 25) and other aspects such as age (26). It is recognized that access to certain technologies is limited for forensic researchers who work on the identification front line. However, it is necessary to consider the possible benefits of technology, such as data sharing without requiring the professional’s physical presence (15, 27–29). This has increased the use of CT images for sex estimation analyses.

The mandible has significant dimorphic points that need to be studied and validated. This is essential for applying different methods, which can support and facilitate the identification process for forensic professionals. Studies show that anthropometric mandible analyses have a high dimorphic potential for correct sex estimation (30–34). However, few Brazilian studies have exclusively used the mandible and three-dimensional (3D) images as a basis for sex estimation, which justifies the importance of this study. Therefore, this study aimed to estimate sex in a sample of Brazilian mandibles.

Materials and Methods

Sample data and study context

This cross-sectional study was initially submitted to the Research Ethics Committee of the University of São Paulo under consolidated approval n. 4.465.406 (CAAE: 40722420.8.0000.0075). The study began after the research had been approved. Its sample consisted of dried mandibles belonging to the collection of the Museum of Human Anatomy of the University of São Paulo, located at the Institute of Biomedical Sciences Hall III (ICB-USP).

The sample consists of an identified osteological collection. At the time of death, data on sex, date of birth, date of death, and racial designation (“skin color”) were recorded by cemetery authorities between 1913 and 1962. The cemeteries from which the sample originated are located in the city of São Paulo, Brazil, a metropolis that currently has more than 12 million inhabitants. Exhumations were carried out after a legally established period had elapsed, during which the family did not claim responsibility for the remains, thus making them available for research purposes (following approval by a Human Research Ethics Committee) (23).

Data collection and eligibility criteria

The study included 93 dry, intact, unfractured mandibles from individuals aged 18 years or older. These mandibles had all anterior incisors preserved and at least two posterior teeth, either premolars or permanent molars. All mandibles from individuals under 18 years old, edentulous mandibles, those that did not meet the criteria for minimum teeth, fractured mandibles, and those with pathologies and malformations were excluded from the sample Figure 1.

Figure 1.

External mandible measures.

The mandibles belonging to the ICB-USP collection underwent CT scans at the University Hospital of the University of São Paulo (HU-USP), using Philips Brilliance 64Ⓡ CT equipment. The CT images comprised the skull and mandible, but they were not articulated when the images were taken. Therefore, the volumetric 3D images of the mandible were subsequently obtained in DICOM format using the Blue Sky Plan 4Ⓡ software. After volumetric reconstruction, the mandibles were isolated, and the skull was excluded from the sample. Then, the 3D images were converted to STL format and saved in a separate file for anthropometric analysis with the MeshmixerⓇ software.

A single, previously calibrated examiner performed all analyses and processes. The intraobserver agreement analysis was performed blindly with an interval of 15 days; the examiner applied the method to 20 mandibles and, after 15 days, the method was reapplied to the same mandibles. The intraobserver error was determined for all 12 measurements applied, individually.

The anthropometric analysis consisted of measuring 12 external anatomical points described by Tunis (30), as shown in Figure 1. The mandibular angle measure was excluded from the study due to its high rate of intraobserver bias. Internal mandible measures were also excluded because this study aimed to analyze only their external measures. The measures were taken using the MeshmixerⓇ software, as shown in Figure 1. All 12 mandibular measures were taken through volumetric analyses of the sample mandibles, which were analyzed from the front, side, and back views, according to the needs of each structure analyzed. All data were entered into a Microsoft Office Excel® spreadsheet.

Data analysis

Quantitative measures were used to analyze the results.

For the analysis of the anthropometric technique applied in this study, Binomial Logistic Regression was used to model the relationship of the binary variable for sex (0 for male and 1 for female) and the interdependent variables (different mandibular measures).

Sensitivity and specificity were calculated using ROC curves based on mandibular measures, determining the area under the curve (AUC) with a 95% confidence interval.

Statistical analyses were conducted using Jamovi® software, version 2.3.28.0 (Computer Software) retrieved from https://www.jamovi.org), considering significance for p-values < 0.05.

Results

Among the 93 mandibles, 56 (60.21%) were male and 37 (39.79%) were female. The mean age of the male sample was 33.85 years, and that of the female sample was 34.40 years. Also, documented background of the sample is described in Table 1.

Table 1. Sociodemographic characteristics of the sample.

		n	%	Mean age (yrs.)
Sex	Male	56	60.21	33.85
	Female	37	39.79	34.40
Documented background	African	42	45.16
	European	30	32.26
	Asian	18	19.35
	No information	03	03.23

Model adjustment measures were obtained for the anthropometric analysis applied in this study, with statistically significant values (p < .001), described in Tables 2 and 3.

Table 2. Model adjustment measures.

Model	Deviation	AIC	R2CS
1	105.74	109,74	0,15

Caption: AIC (Akaike Information Criterion)

Table 3. Model coefficients – ACTUAL SEX (0 = male; 1 = female).

Predictor	Estimates	Standard error	Z	P
Interceptor	10,39	2,78	3,74	<.001
Branch length	-0,18	0,05	-3,88	<.001

Caption: Estimates represent the log odds of "ACTUAL SEX (0 = male; 1 = female) = 1" vs. "ACTUAL SEX (0 = male; 1 = female) = 0"

Results were obtained to measure the sensitivity and specificity of the technique applied to mandibular measures, as shown in Table 4. Among the 12 mandibular measures, the cutoff points ranged from a minimum of 15.63 to a maximum of 113.67. Sensitivity percentages ranged from 55.36% to 80.36%, and specificity percentages ranged from 51.35% to 78.38%. The AUC ranged from 0.61 to 0.76 (Table 4).

Table 4. Measurement of anthropometric properties of mandibular measures.

Measures	Cut-off point	Sensitivity (%)	Specificity (%)	Area under the curve
Ramus length	62.2	71.43	67.57	0.76
Ramus width	31.86	55.36	64.86	0.63
Body length	88.98	60.71	75.68	0.74
Body height P1-P2*	29.59	62.5	78.38	0.74
Body height M2-M3**	23.87	73.21	59.46	0.70
Mandibular angle width	31.53	60.71	78.38	0.73
Coronoid width	19.84	64.29	56.76	0.61
Coronoid height	15.63	75.00	51.35	0.62
Condyle width	17.58	80.36	51.35	0.70
Chin width	16.87	62.5	70.27	0.68
Bicondylar breadth	113.67	75.00	64.86	0.72
Bigonial breadth	91.53	71.43	56.76	0.64

Caption: * P1 (first premolar) / P2 (second premolar) / ** M2 (second molar)/ M3 (third molar).

The sensitivity and specificity values for the different cutoffs demonstrated that the mandibular measures (bicondylar width, body length, body height P1-P2, body height M2-M3, condylar width, mandibular angle width, and ramus length) had percentages above 60% for sensitivity and above 65% for specificity.

The ROC Curve (Figure 2) demonstrates the result of integrating all cutoffs simultaneously with the sensitivity and specificity values, estimating the overall test accuracy. The AUC was above 0.60 in all mandibular measures, with values between 0.61 and 0.76.

Figure 2.

Sensitivity and specificity data obtained through the ROC curve.

Discussion

The Brazilian population exhibits high genetic admixture due to historical immigration waves. This diversity poses a challenge for anthropometric analysis in forensic anthropology, where human identification increasingly relies on digital imaging technologies. Consequently, optimizing these anthropometric techniques has become a key focus in several recent studies (35–37). Forensic anthropology has difficulty in accessing collections and samples and acquiring new pieces due to each country’s ethical and legal issues, which limit maceration processes to obtain osteological pieces (38). Using CT scan images is an alternative, considering the greater ease of access and the fact that physical analysis of the samples is unnecessary, thus preserving collections and pieces for future generations (23).

It is known that CT images can be used for forensic purposes during human identification. Once they are reconstructed in 3D, sex, age, height, and geographic affinity can be estimated, as well as in necessarily physical techniques to approach bone findings (2, 39). The literature portrays the pelvis, skull, and mandible (in this order) as the bone regions with the best results among those with the greatest dimorphic potential (35).

The structures of the mental protuberance and mandible body length are among the measures with the greatest dimorphic potential, as shown in the results of this study.

The mandible, due to its shape and thickness, is one of the bones frequently found in mass disasters and crime scenes. Moreover, it is possible to estimate a person’s biological characteristics through methods that use anthropometric measurement standards at various anatomical points of the mandible (40–42).

Due to its diverse European, African, Asian, and Indigenous origins, the Brazilian population is characterized by a high degree of genetic admixture. Consequently, it is not possible to establish a single, uniform skeletal pattern for this population (15, 43, 44).

Given this context, all mandibles were analyzed collectively, without stratification by geographic location or presumed ancestry, since the extensive admixture that characterizes the Brazilian population makes it difficult to reliably assign individuals to specific ancestral categories based solely on regional or documentary information. Prior classification into ancestral groups could introduce methodological bias and misclassification error, thereby compromising the internal validity of the study. Furthermore, the use of artificial groupings would not accurately reflect the biological complexity of the analyzed collection, which consists of individuals with varying degrees of shared ancestry. Therefore, the decision to perform a pooled analysis aimed to respect population heterogeneity, enhance the statistical robustness of the proposed model, and generate results that are more consistent with the Brazilian forensic reality, in which reliable ancestry information is often unavailable, requiring methods applicable to a highly admixed population.

There are few documented samples available in Brazil, despite the country’s large territorial extent. Therefore, this research was conducted using the sample that was available, and further studies with larger subsamples could better elucidate cut-off points, sensitivity, specificity, and AUC values. Nevertheless, the importance of discussing this issue is substantial and should not discourage research based on the samples available in other parts of the world. Furthermore, for identification purposes in the Brazilian population, it is relevant to include “mixed” data, as this reflects the reality of forensic identification cases. Moreover, the age of the individuals in the sample (older than 18 years) may introduce potential sources of bias, since tooth loss, alveolar resorption, degenerative changes, and bone remodeling are not uniform across this broad age group.

The study by Tunis et al. (30), whose technique we reproduced in the Brazilian sample, used an Israeli population, which has a large biological variability due to immigration to the country, providing a multiracial population with a need for research to study this variability. The study took anthropometric mandible measures, obtaining a total of 25 measures of anatomical points of the mandible, consisting of 13 external and 12 internal measures. All were analyzed through CT images, obtaining concordance rates of up to 90.08% of correctly estimated mandibles in relation to sex.

In contrast, the present study consisted of 12 exclusively external anthropometric measures, aiming to analyze the variability in the results when the internal mandible measures are excluded. The external measurement of the mandibular angle was excluded due to the high rate of intraobserver bias and its lower sexual dimorphic power, as reported by Tunis et al. (30), when analyzing the same measure. The same factor was observed in the present study, and it was decided to exclude this measure.

The cutoff determined in this study showed optimal performance. When analyzing sensitivity and specificity, a positive result was obtained, since the sensitivity percentage ranged from 55.36% to 80.36%, corresponding to the measures of ramus width and condyle width, respectively. Regarding specificity, its percentage ranged from 51.35% to 78.38%, with the measures of coronoid height and condyle width having the lowest percentage, and that of body height p1-p2a and mandibular angle width having the highest percentage of accuracy in estimating sex. The results of this research demonstrate that the measures that correctly estimate sex in the best way are not the same for males and females. The AUC was greater than 0.70 in the following measures: ramus length, body length, height p1-p2a, height m1-m2, mandibular angle width, condyle width, and bigonial width. Hence, these measures obtained statistically acceptable results when estimating sex.

The difficulty in obtaining certain measures in this study is an important factor to be discussed. The chin width is characterized by tactile sensitivity at the time of physical measurement. To achieve greater precision when measuring this width in digital images, incorporating more detailed measurement parameters is essential, as previously demonstrated in a study of modern Black and White South African adults (45).

A comparison with the findings of Tunis et al. (30) reveals that the sex estimation methods yielded a higher accuracy rate in the Israeli population than in the present Brazilian sample. It is worth noting that the present study included a sample of Brazilian individuals, but also mandibles from some South American, North American, European, Asian, and African countries, and Syria, which may be a possible factor for the inferior result. However, other possible reasons can be considered, such as the natural wear of dry mandibles, since the collection has been available for studies for many years.

A study in another Brazilian sample analyzed the volume of the mandibular canal as a parameter for sex estimation, obtaining results of up to 85% agreement (46). A very similar percentage was found in another study using dry mandible CT scans in the Greek population, correctly estimating up to 85.7% of the sample (47). However, it is known that the Greek sample comprises a population with modern Greek characteristics and similar geographic origins, which is a positive factor when estimating sex in a sample.

Limitations of this study include the small sample size, wear and fractures of key anatomical landmarks, the inclusion of edentulous mandibles, and a lack of prior data regarding the biological characteristics of the specimens. However, the collection is one of the most important in Brazil, which has limited collections for study. It is recommended that more methods for estimating sex in the Brazilian population be validated, using anthropological parameters, such as the mandible.

Another limitation that should be discussed is the classification documented by the cemetery and included in the records of the ICB-USP Museum. It was used for comparison with the method applied in the present study. Thus, the only reliable documented data available concern skin color, which constitutes a limitation of the study. This topic has sparked debate among authors regarding the terminology used for geographic origin. In a clarifying review on the subject, Cunha & Ubelaker (2019) state that “reports should focus on the likely ancestry of the examined individual or, alternatively, suggesting how this person likely would have characterized himself/herself or have been socially classified by the communities lived in” (48).

The heterogeneity of the sample is consistent with the population variation observed in Brazil, a country of continental dimensions that currently has more than 213 million inhabitants. Studies conducted in Turkey adopted similar sample grouping strategies, since, despite the possible “intra-sample” variation, the data generated have the potential to assist in real forensic identification cases in that country (49, 50).

Consequently, these results should be interpreted with caution. Phenotypic variation must be considered, as numerous factors can influence sexual dimorphism—including biological ancestry or affinity, population-specific robustness, diet, functional loading, dental wear, dentition status, advanced age, post-tooth loss bone remodeling, and secular changes. Further studies are needed to clarify these issues.

In conclusion, the results of this study highlight the importance of investigating distinct populations and utilizing samples from diverse skeletal collections in forensic anthropology. Because studies utilizing CT images of dry mandibles remain scarce in Brazil, expanding this line of research is essential, especially given the vast number of collections available in the country. Digitizing collections is also a relevant topic, and some measures are difficult to perform digitally. Therefore, a booklet with clear guidelines for mandibular measurement in CT images would be useful. Moreover, internal measures of the mandible should be taken together with external ones.

Conclusion

In conclusion, tomographic-based mandibular anthropometric analysis serves as a valuable complementary tool for forensic sex determination, particularly in forensic cases where complete skeletal material is unavailable or fragmented.