Deep Learning for Age Estimation and Sex Prediction Using Mandibular-Cropped Cephalometric Images: Comparative Model Development and Validation Study.

Background: Mandibular structures offer resilient features for forensic identification where partial remains are available in postmortem condition. Deep learning applied to cephalometric radiographs offers an opportunity to predict demographic attributes, such as age and sex, which are critical in forensic and clinical contexts. Objective: This study aimed to develop and evaluate a multitask deep learning framework for age estimation and sex prediction from cropped mandibular regions of cephalometric radiographs, comparing multiple convolutional neural network backbones and preprocessing scenarios to address class imbalance. Methods: A total of 340 anonymized cephalometric radiographs from Indonesian individuals aged 8 to 40 years were collected and manually cropped into 2 mandibular regions of interest: mandibular length and mandibular angle, producing 680 validated samples. Images were resized to 224×224 pixels and processed under 4 preprocessing scenarios: original, Synthetic Minority Oversampling Technique, StandardScaler, and Synthetic Minority Oversampling Technique+StandardScaler. Six pretrained convolutional neural network backbones (MobileNetV2, ResNet50V2, InceptionV3, InceptionResNetV2, VGG16, and VGG19) were fine-tuned within a multitask framework. Performance was evaluated using mean absolute error and mean absolute percentage error for age estimation and accuracy and F1-score for sex prediction. Results: VGG16 achieved the best performance for age estimation, with the lowest mean absolute error of 3.19 years and mean absolute percentage error of 13.19% in the original dataset. For sex prediction, VGG16 achieved the highest accuracy (86%) and balanced F1-scores (female: 92%; male: 63%) under the StandardScaler condition, followed by VGG19 (accuracy=82%). Conclusions: Combining mandibular cropping with deep learning and balanced preprocessing scenarios enhances demographic prediction in cephalometric radiographs. The findings emphasize the potential use of artificial intelligence-assisted forensic odontology to support disaster victim identification when partial remains are available.