Scoping review maps CT evidence on sinuses for sex, age, stature estimation

A sweeping scoping review has mapped the entire published evidence base on using CT and cone-beam CT measurements of the paranasal sinuses and craniofacial bony landmarks to estimate a person’s sex, age, or stature — and it has exposed both a maturing field and some striking blind spots. The review, published in the Egyptian Journal of Forensic Sciences, identified 76 eligible studies from 18 countries, spanning research from database inception to 7 May 2026. Its central verdict: sex estimation from these structures is now well established, age estimation is far weaker, and stature estimation has never been studied at all within the eligible literature.

The review team, led by Hevin Ujwal Vas and B Suresh Kumar Shetty, searched PubMed, Scopus, and Web of Science and screened records independently in duplicate, achieving an inter-rater agreement of Cohen’s kappa = 0.87, which the authors describe as almost perfect. They structured the review around the Population, Concept, Context framework: human subjects undergoing cranial or paranasal CT-based imaging, CT-based craniofacial and sinus measurements, and forensic estimation of sex, age, or stature. Following the Arksey and O’Malley methodology and Joanna Briggs Institute guidelines, and reporting against the PRISMA-ScR checklist, they charted data from all 76 included studies without pooling effect estimates — a deliberate choice, since the aim was to map the extent and nature of the literature rather than judge the certainty of any single quantitative effect.

The geographic distribution of the evidence is heavily skewed. India leads with 24 studies, followed by Brazil with 11 and Iran with 8; Egypt and Turkey contributed 7 studies each and China 5. The remainder came from single-digit contributions across Romania, Saudi Arabia, Korea, Indonesia, Italy, Sri Lanka, Nigeria, the United States, Jordan, France, Czechia, and Japan. Most studies used retrospective or cross-sectional designs, with sample sizes ranging from just 20 participants to 1,018. Cone-beam CT was the dominant imaging modality, ahead of conventional, multidetector, or multislice CT. Publication output has accelerated dramatically: only isolated papers appeared between 2012 and 2016, but 44 of the 76 studies — 57 percent — were published in just the last three years, a surge the authors attribute to wider availability of CBCT imaging and growing use of machine learning in forensic anthropology.

Anatomically, one structure dominates. The maxillary sinus was examined in 55 of the 76 studies, or 72 percent, reflecting its accessibility, size, and relatively consistent morphology. The frontal sinus, prized for its highly individual-specific morphology, appeared in 33 studies (43 percent), and the sphenoid sinus in 14 (18 percent). The ethmoid sinus was markedly neglected, featuring in only a single study. A smaller subset of work measured non-sinus hard-tissue landmarks — bizygomatic breadth, interorbital breadth, nasion-related distances, and composite craniometric sets — sometimes alone and sometimes combined with sinus measurements to test whether merging the two improves classification accuracy.

The outcome distribution is even more lopsided. Sex estimation was assessed in 72 of 76 studies, or 94 percent of the corpus. Age estimation appeared, alone or alongside sex, in 16 studies (21 percent), most of which tackled both outcomes in the same sample. Stature estimation — despite being one of the four classical pillars of the biological profile alongside sex, age, and ancestry — was addressed by no study in the corpus. The authors call this a clear and complete evidence gap, noting that many of the same CT and CBCT datasets already used for sex and age estimation could, in principle, support stature analyses if height data were simply recorded.

Methodologically, classical discriminant function analysis remains the standard statistical approach, often used alone or alongside logistic regression. But a growing minority of recent studies, particularly from 2023 onward, have applied machine learning and artificial neural network methods, including random forest, support vector machines, k-nearest neighbours, and multilayer perceptron or deep learning classifiers. Several of the most recent studies reported that machine learning classifiers matched or outperformed conventional discriminant analysis, and combining multiple sinuses or sinus-plus-craniometric variables generally improved accuracy over single-structure models. The authors caution, however, that machine learning studies in this corpus were typically single-population and internally validated, so the same generalisability concerns apply as to classical methods.

Reported sex-classification accuracy varied enormously, from roughly 60 to 65 percent for single linear indices such as the frontal sinus index used alone, up to the low-to-mid 90s for combined multi-sinus models, geometric morphometric approaches, and machine learning classifiers. A handful of very small single-centre studies reported accuracies at or above 96 to 100 percent — figures the review urges readers to interpret cautiously given limited sample sizes and the general absence of external validation. Among the more specific findings: combining 11 craniofacial distances in a logistic regression model reached 84.8 percent accuracy for sex determination, compared with 83.6 percent using discriminant analysis, while cranial metrics alone reached 77 percent. Bilateral maxillary sinus height, width, and depth combined in a discriminant function model reached 88.4 percent. Combining maxillary, frontal, and sphenoid sinus volumes in a discriminant analysis reached 95.7 percent, with frontal sinus volume alone reaching 94.4 percent. Several studies also reported a systematic sex asymmetry in classification performance, with female cases classified more accurately than male cases, or vice versa, depending on the population and structure examined.

The age-estimation evidence is thinner and less consistent. Where examined, sinus volume and linear dimensions generally showed weak-to-moderate negative correlations with advancing age at 40 years and above, consistent with known sinus pneumatisation changes across the lifespan. But several studies explicitly concluded that sinus measurements alone are not reliable for age estimation, particularly in adults over roughly 40 years. Regression-based age models generally explained only a modest proportion of variance, and mean absolute errors, where reported, widened substantially in older age groups. In short, the review finds that sinus measurements are not yet established as dependable age estimators, especially beyond the fourth decade of life.

Perhaps the most consequential theme is population specificity. Study authors repeatedly acknowledged that sinus and craniofacial dimensions are population-, sex-, and age-specific, and that discriminant functions derived in one population frequently do not generalise well to others. Relatively few studies performed external, cross-population or cross-sample validation of their models; most relied on internal cross-validation or a single training and testing split within the same population. The review’s authors stress that accuracy figures in the 90s should be read with this in mind, since they typically come from small, single-population samples tested on the same data used to build the model, and cannot be directly compared across studies without accounting for differences in sample size, structure combinations, sex ratio, age range, and validation design.

For forensic practitioners, the review supports using paranasal sinus and craniofacial measurements as a supplementary line of evidence for sex estimation, particularly when combined across multiple structures — but warns against applying published formulae outside the population in which they were derived without local validation. For researchers, the authors set out four priorities: launching the first stature-estimation studies using these structures, testing existing sex- and age-estimation models on independent samples from other populations, standardising landmark definitions and measurement protocols so data can eventually be pooled, and evaluating machine learning models on held-out or external test sets rather than internal cross-validation alone. The review also acknowledges its own limits: the search covered only three databases, only English-language studies with accessible full texts were eligible, and most included studies came from single institutions without reporting ethnic composition — meaning some relevant research may be missing and representativeness could not be assessed. Closing these gaps, the authors conclude, is what would turn a broad but fragmented literature into evidence that forensic practitioners can confidently apply beyond the populations in which it was generated.

Subject of Research: CT-based craniofacial and paranasal sinus measurements for forensic estimation of sex, age, and stature

Article Title: CT-based hard-tissue craniofacial landmarks and paranasal sinuses for biological profile estimation (sex, age and stature) in human populations: a scoping review

Article References: Vas, H. U., Shetty, B. S. K., Shetty, P., Rai, S., Mithal, D., Balbudhe, M. S., & Render, R. (2026). CT-based hard-tissue craniofacial landmarks and paranasal sinuses for biological profile estimation (sex, age and stature) in human populations: a scoping review. Egyptian Journal of Forensic Sciences, 16(1), Article 83. https://doi.org/10.1186/s41935-026-00582-1

Image Credits: AI Generated

DOI: 10.1186/s41935-026-00582-1

Keywords: forensic anthropology, scoping review, computed tomography, paranasal sinuses, sex estimation, age estimation, stature estimation, discriminant function analysis, machine learning, CBCT, craniofacial landmarks, external validation