Age estimation from pubic symphysis based on cinematic volume rendering: comparison between Suchey-Brooks staging and deep learning.
Authors
Affiliations (8)
Affiliations (8)
- Department of Forensic Pathology, West China School of Basic Medical Sciences & Forensic Medicine, Sichuan University, Chengdu, 610041, Sichuan, P.R. China.
- Forensic Science Services of Sichuan Provincial Public Security Department, Chengdu, 610504, Sichuan, China.
- College of Computer Science, Sichuan University, Chengdu, 610041, Sichuan, P.R. China.
- Shanghai Key Laboratory of Forensic Medicine, Shanghai Forensic Service Platform, Academy of Forensic Science, Ministry of Justice, Shanghai, 200063, China.
- Center of Forensic Expertise, Affiliated hospital of Zunyi Medical University, Zunyi, Guizhou, 563000, China.
- College of Computer Science, Sichuan University, Chengdu, 610041, Sichuan, P.R. China. [email protected].
- Department of Forensic Pathology, West China School of Basic Medical Sciences & Forensic Medicine, Sichuan University, Chengdu, 610041, Sichuan, P.R. China. [email protected].
- Department of Forensic Pathology, West China School of Basic Medical Sciences & Forensic Medicine, Sichuan University, Chengdu, 610041, Sichuan, P.R. China. [email protected].
Abstract
Accurate adult age estimation remains a central challenge in forensic anthropology. This study compared the performance of the conventional Suchey-Brooks (SB) method with a deep learning (DL) approach in adult age estimation using standardized cinematic volume rendering (cVR) from pelvic CT scans. A total of 1, 359 examinations from a Chinese cohort were analyzed. SB-based cubic regression achieved mean absolute errors (MAE) of 5.94 years in males and 6.06 years in females, while the DL model produced MAEs of 6.64 and 7.03 years, respectively. No significant accuracy difference was observed between methods. Both exhibited age-dependent bias, with overestimation in younger adults and underestimation in older individuals. Visualization using gradient-weighted class activation mapping indicated that the DL model focused on key morphological features of the pubic symphyseal surface, with patterns varying across age groups and sex. These results demonstrate that both cVR-based SB phase assignment and deep learning regression achieve compatible precision, with deep learning providing a scalable, automated, first-line approach for objective age estimation.