Machine learning-driven forensic sex prediction using CT-based nasal and maxillary sinus metrics.
Authors
Affiliations (9)
Affiliations (9)
- Department of Forensic Medicine and Clinical Toxicology, Faculty of Medicine, Benha University, Benha, Egypt.
- School of Science, Faculty of Engineering and Science, University of Greenwich (Medway Campus, Kent), London and South East University Group, London, UK.
- Department of Zoonoses, Faculty of Veterinary Medicine, Zagazig University, Zagazig, Egypt.
- Department of Diagnostic Radiology, Faculty of Medicine, Assiut University, Assiut, Egypt.
- Department of Diagnostic and Interventional Radiology, Faculty of Medicine, Benha University, Benha, Egypt.
- Department of Family & Community Medicine, Faculty of Medicine, Northern Border University, KSA, Arar, Saudi Arabia.
- Department of Community, Environmental, and Occupational Medicine, Faculty of Medicine, Benha University, Benha, Egypt.
- Faculty of Computers and Artificial Intelligence, Cairo University, Giza, Egypt.
- Department of Forensic Medicine and Clinical Toxicology, Faculty of Medicine, Assiut University, Assiut, Egypt. [email protected].
Abstract
Sex determination is a key component of forensic identification, especially in cases involving fragmented or decomposed human remains where conventional skeletal markers are unavailable. The maxillary sinus and nasal structures are protected craniofacial components that exhibit sexual dimorphism and may aid sex estimation. This study evaluated computed tomography (CT)-based anthropometric measurements of these structures for sex determination in an Egyptian population using machine learning (ML). A comparative cross-sectional study was conducted on 195 adult Egyptians (100 females, 95 males) from Upper Egypt (Assiut, n = 104) and Lower Egypt (Benha, n = 91). CT scans of the paranasal sinuses were analyzed to obtain six maxillary sinus dimensions and three nasal measurements. Age and ten additional engineered features were derived, yielding 20 features for ML analysis. Two ML frameworks differing in the sequence of feature selection and hyperparameter optimization were evaluated using six classifiers and five feature-selection methods. Performance was assessed using accuracy, area under the receiver operating characteristic curve (AUC), precision, recall, F1-score, and specificity. Significant sex-related differences were observed in several measurements, with regional variation between Upper and Lower Egypt. Framework 2 generally outperformed Framework 1. The best-performing models achieved AUCs of 0.771 and 0.768, while the highest accuracy reached 74.4%. Nasofrontal angle, nasion-tip distance and mean anteroposterior maxillary dimension were the most consistent predictors. CT-based nasal and maxillary sinus anthropometry shows moderate utility for forensic sex estimation in Egyptians. ML improved classification by capturing complex morphometric relationships, with nasal measurements emerging as robust sex-discriminative markers.