Weighted estimation of multiple regions in CT-derived images of the pubic symphysis for age estimation using a deep learning framework with R-Mixup augmentation
6 affiliations
- 1Shanghai Key Laboratory of Forensic Medicine and Key Laboratory of Forensic Science, Academy of Forensic Science, Ministry of Justice
- 2Institute of Forensic Science, Fudan University
- 3Fujian Hengxin Judicial Identification Center
- 4School of Computer Engineering and Science, Shanghai University
- 5School of Forensic Medicine, Shanxi Medical University
- 6Forensic Laboratory, Public Security Bureau of Changzhou
Abstract
Deep learning has been widely used for forensic age estimation based on CT-derived images, and data volume is critical to enhancing the model performance. However, traditional data augmentation such as flipping, rotating, and adjusting contrast is not well-suited for regression tasks like age estimation. This study applied a data augmentation, R-Mixup to DenseNet121 training for age estimation from CT-derived images of the pubic symphysis. A dataset of 1,200 individuals aged 20–80 years was analyzed, with images segmented into five regions of interest (ROIs): pubic symphysis surface (PSS), ischial tuberosity (IT), pubic dorsal aspect (PDA), pubic ventral aspect (PVA) and pubic tubercle (PT). Results show that R-Mixup significantly enhanced model performance, particularly when combined with transfer learning, reducing estimation MAE errors for all ROI images. PVA was the most informative ROI, achieving the lowest MAE of 4.89 years for females and 6.32 years for males. Weighted estimation based on multiple ROIs further improved accuracy, with the best-performing combinations reaching MAE of 4.55 years (females) and 6.00 years (males). This study demonstrates the effectiveness of CNN-based forensic age estimation, highlighting the benefits of multi-ROI analysis and non-conventional data augmentation. Future research will explore co-training techniques to improve model robustness in cases involving incomplete skeletal remains.
Related
- IFC-HFlowVAE: A self-enhancing generative framework with structural anchoring for imbalanced clinical data augmentation
- Diagnostic accuracy, time window estimation, and prognostic value of diffusion-weighted MRI in acute ischemic stroke: a systematic review and meta-analysis
- Marking-based load balancing Weighted Clustering Framework for Mobile ADHOC Network
- A privacy-preserving multicenter framework for lung nodule segmentation in low-dose CT using an F1-score-weighted federated 3D Transformer-UNet
- EPRA U-Net: an efficient pyramid residual attention framework for accurate infarct segmentation in diffusion-weighted MRI.
BibTeX
@article{lyu2026weighted,
title = {{Weighted estimation of multiple regions in CT-derived images of the pubic symphysis for age estimation using a deep learning framework with R-Mixup augmentation}},
author = {Zhihe Lyu and Mingzhen Yang and Yongjie Cao and Yihua Zhao and Zhihan Zuo and Huaihan Zhang and Weiwei Zhu and Yuchun Fang and Ping Huang and Ji Zhang},
journal = {Scientific Reports},
year = {2026},
month = jul,
doi = {10.1038/s41598-026-62705-x},
url = {https://doi.org/10.1038/s41598-026-62705-x},
note = {CC BY},
}
RIS
TY - JOUR
TI - Weighted estimation of multiple regions in CT-derived images of the pubic symphysis for age estimation using a deep learning framework with R-Mixup augmentation
AU - Zhihe Lyu
AU - Mingzhen Yang
AU - Yongjie Cao
AU - Yihua Zhao
AU - Zhihan Zuo
AU - Huaihan Zhang
AU - Weiwei Zhu
AU - Yuchun Fang
AU - Ping Huang
AU - Ji Zhang
PY - 2026
DA - 2026/07/01
JO - Scientific Reports
DO - 10.1038/s41598-026-62705-x
UR - https://doi.org/10.1038/s41598-026-62705-x
AB - Deep learning has been widely used for forensic age estimation based on CT-derived images, and data volume is critical to enhancing the model performance. However, traditional data augmentation such as flipping, rotating, and adjusting contrast is not well-suited for regression tasks like age estimation. This study applied a data augmentation, R-Mixup to DenseNet121 training for age estimation from CT-derived images of the pubic symphysis. A dataset of 1,200 individuals aged 20–80 years was analyzed, with images segmented into five regions of interest (ROIs): pubic symphysis surface (PSS), ischial tuberosity (IT), pubic dorsal aspect (PDA), pubic ventral aspect (PVA) and pubic tubercle (PT). Results show that R-Mixup significantly enhanced model performance, particularly when combined with transfer learning, reducing estimation MAE errors for all ROI images. PVA was the most informative ROI, achieving the lowest MAE of 4.89 years for females and 6.32 years for males. Weighted estimation based on multiple ROIs further improved accuracy, with the best-performing combinations reaching MAE of 4.55 years (females) and 6.00 years (males). This study demonstrates the effectiveness of CNN-based forensic age estimation, highlighting the benefits of multi-ROI analysis and non-conventional data augmentation. Future research will explore co-training techniques to improve model robustness in cases involving incomplete skeletal remains.
SN - 2045-2322
ER -