Artificial intelligence in forensic sex estimation: A narrative review of machine learning and deep learning approaches across skeletal regions

Abstract

Introduction: With the development of artificial intelligence (AI ), especially machine learning (ML) and deep learning (DL) techniques, numerous studies have emerged that apply these methods to sex estimation based on metric and morphometric features of skeletal bones. This narrative review aims to summarise and evaluate machine learning and deep learning methods in sex estimation based on the measurements and morphometry of the pelvis, skull, long bones, and hyoid bone. Methods: A detailed search of ML and DL techniques in sex estimation from the skeletal regions was conducted through PubMed and Scopus databases from January 2015. After the final selection, original articles, systematic reviews, and metaanalyses were analysed. Results: ML models based on morphometric measurements of the pelvis achieved accuracy rates of 85–97.33% in various populations. DL methods, particularly Convolutional Neural Networks, eliminating the need for manual measurements, achieved accuracy rates of 99.8% for the pelvis, over 96% for the cranium, and 89% for long bones. For ML models using the hyoid bone, the accuracy rates were 69.2–95.4%, and for DL models, up to 89%. These systems provide automated sex identification with reproducible, objective, and rapid assessment. However, critical limitations persist, including data dependency and population bias, limited explainability, legal and ethical uncertainty, impracticality, and the absence of international validation standards. Conclusions: The current state of knowledge and level of technological development indicate that AI should serve as a support tool rather than a standalone replacement for expert anthropological assessment. Its role is to enhance repeatability, minimise bias, and clarify confidence levels in skeletal sex estimation.