Aug 2026· Zeitschrift für Rechtsmedizin· 0 citations· 30 references
Medicine
TL;DR
Both YOLOv8 and YOLOv11 outperformed YOLOv5 across all metrics, demonstrating superior feature extraction capacity and robust convergence and indicate that next-generation YOLO architectures enable highly accurate and fully automated sex estimation from hand-wrist radiographs without the need for manual annotation or region-of-interest selection.
Sex estimation by forensic investigators can be enhanced through the use of artificial intelligence, which has shown considerable promise in the interpretation of medical images. This study introduces a novel multimodal deep learning framework that combines chest radiographic features with manually annotated anatomical landmarks, enabling more accurate sex prediction. This retrospective study included chest radiographs from 563 Egyptian patients obtained from the Department of Diagnostic and Interventional Radiology, Faculty of Medicine, Cairo University. A custom deep convolutional neural network was developed comprising two parallel branches: (1) an image-processing branch utilizing residual convolutional blocks with batch normalization, skip connections, max pooling, global average pooling, and fully connected layers; and (2) an anatomy branch processing 45 structured anatomical features derived from one JSON annotation file. Features from both branches were fused through concatenation and classified using fully connected layers with dropout regularization and a sigmoid output layer. The model achieved an overall accuracy of 83% and Area under the curve = 0.8776. Class-wise performance showed a precision, recall, and F1-score of 0.79, 0.89, and 0.84, respectively, for female samples, and 0.82, 0.68, and 0.75, respectively, for male samples. Upon further study of the results, sex classification accuracy reached 90% when anatomical landmark annotations were incorporated into the prediction process. Although the model learned anatomical relationships during training, some predictions were generated without utilizing this anatomical information and resulted in a lower accuracy of 71%. The proposed multimodal framework provides a reliable approach for sex estimation by combining chest radiographs with structured anatomical annotations. The findings highlight the importance of annotation-guided models in forensic identification and support the development of large-scale national radiographic datasets to facilitate the implementation of artificial intelligence-assisted medico-legal assessments, particularly in scenarios requiring positive identification such as disaster victim identification.
K. Kamel, Doaa Tawfik, Nashwa Mohammed Saged et al.· Egyptian Journal of Forensic...· 0 citations
Purpose: To develop and validate a deep learning ensemble for estimating adult sex, age, height, and weight from coronal digitally reconstructed radiographs (DRRs) generated from diagnostic CT. Materials and Methods: This retrospective study included 128,621 CT examinations from 80,004 adults at nine institutions in Japan. Three multitask models-ConvNeXt-Base, ViT-Base/16, and MaxViT-Base-were fine-tuned using coronal DRRs and combined by weighted averaging. Data were split by institution into training (114,147 examinations; seven institutions), tuning (4,305; one institution), and test (10,169; one institution) sets; generalizability was assessed on two non-Japanese datasets. Accuracy and mean absolute error (MAE) were used to evaluate sex classification and age, height, and weight regression, respectively. Body surface area (BSA)-corrected heart and liver volume trends were compared using true versus estimated height and weight. Results: In the test set (median age, 69.9 years; 4,899 of 10,169 [48.2%] male), overall sex-classification accuracy was 0.997 (95% CI, 0.996-0.998), and MAEs were 3.57 years (3.51-3.63), 2.59 cm (2.54-2.64), and 3.40 kg (3.34-3.47) for age, height, and weight, respectively. In examinations covering the chest through pelvis, accuracy was 1.000, and MAEs were 3.15 years, 2.28 cm, and 3.18 kg, respectively. BSA calculated from estimated values reproduced age-related heart and liver volume trends obtained using true values. On non-Japanese datasets, height error increased but was reduced by continued fine-tuning. Conclusion: The ensemble estimated adult sex, age, height, and weight from CT-derived DRRs, with generally lower errors in examinations with broader anatomical coverage.
Tomohiro Kikuchi, Kohei Yamamoto, Y. Nomura et al.· arXiv.org· 0 citations
OBJECTIVE
This study evaluated the feasibility and performance of a deep learning-based multitask framework for estimating sex and age using femoral CT images in a Japanese population.
MATERIALS AND METHODS
Postmortem CT (PMCT) data from 1,489 Japanese individuals (1,003 males, 486 females) aged 20-90 years were analyzed. Femora were automatically segmented using the TotalSegmentator library, and three-dimensional volumetric data were processed using a three-dimensional Vision Transformer foundation model (3DINO) adapted with low-rank adaptation. The model was trained in a multitask learning framework to perform sex classification and age estimation simultaneously. Performance was assessed, and attention map analysis was conducted to identify anatomically relevant regions contributing to predictions.
RESULTS
On the independent test dataset, sex estimation showed high performance, with balanced accuracies of 0.897 (95% confidence interval [CI], 0.887-0.948) for the left femur and 0.929 (95% CI, 0.907-0.961) for the right femur. Area under the curve values were also high (0.979 and 0.988 for the left and right femora, respectively). Age estimation achieved a mean absolute error of approximately 8 years and a root mean square error of approximately 10 years; female models showed lower mean absolute error and root mean square error values than male models. Attention map analysis indicated that the distal femur contributed most to sex estimation, whereas the femoral shaft was the primary region associated with age estimation.
CONCLUSION
A foundation model-based approach enables accurate simultaneous estimation of sex and age using femoral CT images and has potential applications in automated forensic biological profiling.
Unknown authors· Zeitschrift für Rechtsmedizi...· 0 citations
Automated analysis of panoramic radiographs remains challenging due to anatomical complexity and image variability. While deep learning has shown strong performance in dental imaging, most studies focus on isolated tasks. This study aimed to propose a hierarchical YOLOv8-based framework aligned for comprehensive analysis of panoramic radiographs using structured dental annotations. A three-stage deep learning pipeline based on YOLOv8 was developed using the DENTEX dataset. The framework includes (1) quadrant classification, (2) tooth enumeration (FDI 11-48), and (3) tooth-level abnormality detection using a two-stage approach (binary screening followed by subtype classification). Panoramic radiographs with hierarchical annotations were used, with an 80:20 train–validation split. Performance was evaluated using mAP50, mAP50-95, precision, recall, and F1-score. The model achieved near-perfect performance for quadrant classification (mAP50=0.994, mAP50-95=0.750, precision=0.994, recall=0.995, and F1=0.994) and strong performance for tooth enumeration (mAP50=0.936, mAP50-95=0.536, precision=0.902, recall=0.897, and F1=0.899). Abnormality detection showed moderate performance (mAP50=0.687, mAP50-95=0.480, precision=0.655, recall=0.742, and F1=0.696). At the class level, impacted teeth (F1=0.904) and caries (F1=0.885) were well detected, whereas periapical lesions (F1=0.568) and deep caries (F1=0.585) showed lower performance. Precision and recall were balanced across tasks. The proposed hierarchical framework enables anatomical localization and integration of detection tasks of panoramic radiographs within a unified pipeline using YOLOv8. While performance is near-ceiling for anatomical tasks, disease detection remains challenging, particularly for low-contrast lesions.
Ho-Ang Tung, Young-Seok Park, Khoa Van Pham et al.· Odovtos International Journa...· 0 citations
Forensic odontology plays a crucial role in human identification, particularly in estimating age and sex in individuals without official identification documents. Although traditional methods are widely used, they have significant limitations such as population variability, observer-related errors, and low reproducibility. In recent years, artificial intelligence (AI) and machine learning approaches have provided more objective, rapid, and reproducible alternatives. In particular, convolutional neural network-based algorithms have reduced mean absolute error in age estimation and achieved accuracy rates exceeding 90% in sex classification by automatically extracting discriminative features from panoramic radiographs and cone-beam computed tomography data. National and international studies support the reliability of these technologies, especially in pediatric and adolescent populations where traditional methods are less precise. However, dataset biases, population-specific variability, limited generalizability across ethnic groups, and the non-transparent nature of deep learning models continue to pose challenges for interpretability and legal admissibility. Therefore, ethical and regulatory frameworks emphasizing transparency, data protection, and explainable-AI (XAI) principles are essential for the responsible implementation of these technologies. Future progress will depend on large-scale, multicenter validation studies and the integration of multimodal datasets combining radiographic, morphological, and demographic information. The adoption of XAI frameworks is expected to enhance transparency, accountability, and forensic reliability, thereby enabling. AI-based systems to become scientifically robust and legally valid tools in forensic odontology.
Burak Çarıkçıoğlu· Turkish Journal of Forensic...· 0 citations
Objectives: The accurate quantification of hip deformities in medical images presented a significant challenge for radiologists. Inaccurate quantifications can lead to misclassification of diseases and improper treatment. This study introduced two-stage deep learning (DL) models designed for the detection of landmarks in hip X-ray images, aiming to improve the accuracy of deformity quantification. Methods: This study employed two two-stage deep learning (DL) models based on VGG16 and ResNet50. The first stage aimed to detect the bounding box for each landmark, while the second stage focused on pinpointing the exact location of the landmark within the magnified box. The model automatically identified 16 hip landmarks and measures 12 specific dimensions on each side. Training, validation, and testing were conducted on a dataset comprising 854 3-joint lower limb X-ray images depicting various anomalies. Results: The model's measurements were compared with those obtained manually to evaluate performance. The resulting average error ranged from 0.56 to 2.33 mm for different landmarks in the ResNet50-based model. The most accurate measurements were obtained for the femoral head radius and shaft width, with average errors of 0.49 mm and 0.64 mm, respectively. Conversely, the least accurate results were observed for the alpha angle, with an error of 7.37°. Conclusion: This dataset can be valuable for diverse research endeavors related to the hip region. The 2-stage model demonstrated a brief learning time due to the small image size loaded in both steps, while also offering high precision through the use of a bounding box.
Unknown authors· The Archives of Bone & Joint...· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
