BIOLOGICAL SEX DETERMINATION IN CADAVERS USING DEEP LEARNING ALGORITHMS FROM COMPUTED TOMOGRAPHY IMAGES OF PELVIS AND SKULL

Research Question and Approach

This study investigates the feasibility of using deep learning algorithms to automate the determination of biological sex from post-mortem computed tomography (PMCT) scans. Traditional forensic anthropology relies on manual, observer-dependent morphological analysis, which can be difficult in cases of decomposition or trauma. The researchers evaluated seven state-of-the-art deep learning architectures—including YOLO26, YOLO11, ConvNeXt-Tiny, EfficientNetV2, ViT-B16, VGG16, and ResNet50—using a dataset of 141 autopsied cadavers. To optimize performance, the team converted 3D volumetric CT data into standardized 2D profile projections and employed transfer learning to compensate for the limited sample size.

The best-performing model, YOLO26, demonstrated high consistency in sex classification, particularly when analyzing the pelvic region. The model achieved an overall patient-level accuracy of 95.65%, a recall of 92.86%, and an F1-score of 94.36%. The researchers found that the model maintained robust performance even in the presence of trauma-related artifacts and anatomical disfigurements, suggesting that deep learning can provide an objective and high-speed alternative to traditional manual skeletal analysis.

Automated sex determination has significant implications for forensic investigations, especially in mass disaster victim identification or cases where remains are fragmented or decomposed. By reducing reliance on subjective human interpretation, these AI-driven methods offer a more reproducible and efficient workflow for forensic pathologists. The success of this methodology indicates that even with limited datasets, deep learning can effectively learn the subtle sexual dimorphism present in skeletal structures, provided that appropriate data augmentation and dimensionality reduction techniques are applied.