RESUMO
Objective Explore the feasibility and accuracy of using deep learning techniques for gender inference in panoramic dental radiography images of Chinese Han population.Methods A total of 10,600 OPG images from Han individuals aged 18 to 70(5,300 males and 5,300 females)were collected and randomly divided into training set,validation set,and test set in an 8:1:1 ratio.MobileNetV2,Swin Transformer Small,and Swin Transformer Tiny models were trained,and the classification performance of the models was evaluated and visually displayed using accuracy,F1 score,and Grad-CAM algorithm.Results The accuracy of MobileNetV2,Swin Transformer Small,and Swin Transformer Tiny models was 97.57%,95.13%,and 96.28%respectively,with MobileNetV2 model showing the best overall performance.The Grad-CAM algorithm revealed that male OPG images mainly focused on the left and right mandibular branches and alveolar bone,while female OPG images mainly focused on the maxillary sinus,left mandibular branch,and posterior alveolar bone.Conclusion This study demonstrates that the gender inference model based on deep learning techniques for OPG images of Chinese Han population has high accuracy and generalization ability,providing a new approach for forensic gender determination in forensic medicine.
