Sex prediction based on computed tomography images of the mandible using deep learning models
Küçük Resim Yok
Tarih
2026
Dergi Başlığı
Dergi ISSN
Cilt Başlığı
Yayıncı
Taylor & Francis Ltd
Erişim Hakkı
info:eu-repo/semantics/closedAccess
Özet
The aim of this study is to estimate sex using deep learning methods from mandibular images obtained from computed tomography (CT) scans. In this study, 2310 images were recorded in jpeg format by segmenting the superior, inferior, anterior, posterior, right side, and left side of the mandible from retrospective and randomly scanned CT images belonging to 184 women and 201 men aged 18-65 years. The obtained data were divided into an 80% training set and a 20% test set, and the performance of the deep learning methods ConvNetBase, InceptionV3, Data-Efficient Transformer (DeiT), and the proposed hybrid model were evaluated and compared. In the study, the hybrid model was found to be the most successful model with a 92.50% accuracy rate, 0.0750 lowest error rate (MAE), 92.48% F1-score value, and 0.95 AUC-ROC value. In terms of accuracy, the hybrid model was followed by InceptionV3 (92.17%), ConvNetBase (88.67%), and DeiT (86%). In the Gradient-weighted Class Activation Mapping (Grad-CAM) analysis, it was determined that the middle and lower mandibular regions contributed significantly to sex prediction. We hope that the hybrid model, which showed a high accuracy rate in our study, will guide forensic doctors and anthropologists in sex prediction.
Açıklama
Anahtar Kelimeler
[Keyword Not Available]
Kaynak
Australian Journal of Forensic Sciences
WoS Q Değeri
Q4
Scopus Q Değeri
Q3
