Estimation of population affinity from computed tomography-derived pelvic measurements: a comparative study of Indonesian and Japanese population samples.
Authors
Affiliations (5)
Affiliations (5)
- Centre for Forensic Anthropology, School of Social Sciences, University of Western Australia, Crawley, WA, 6009, Australia. [email protected].
- Department of Forensic Medicine, Graduate School of Medicine, The University of Tokyo, Tokyo, 113-0033, Japan. [email protected].
- Centre for Forensic Anthropology, School of Social Sciences, University of Western Australia, Crawley, WA, 6009, Australia.
- Department of Forensic Medicine, Graduate School of Medicine, The University of Tokyo, Tokyo, 113-0033, Japan.
- Radiology Department, Hasanuddin University, Jalan Perintis Kemerdekaan KM. 10, Talamanrea, Makassar, 90254, Indonesia.
Abstract
Population affinity estimation is a core component of forensic anthropology, particularly in cases where DNA, dental records, and fingerprint evidence are unavailable. Although cranial morphology has been extensively investigated, direct comparative studies of pelvic morphology among Asian populations remain scarce. This study quantified pelvic morphometric differences between contemporary Indonesian and Japanese adults and evaluated the feasibility of population affinity classification using measurements derived from computed tomography (CT). CT scans of 185 Indonesian and 236 Japanese adults were analyzed. Eleven pelvic measurements were obtained from three-dimensional reconstructions, and interpopulation differences were assessed using nonparametric statistical tests. Machine learning models, including random forest and support vector machines, were developed for sex-mixed binary classification, sex-specific binary classification, and four-way classifications by population and sex. Significant differences were evident across all measurements. Japanese individuals exhibited consistently larger mean values than Indonesians of the same sex, while patterns of sexual dimorphism were comparable between populations. Classification accuracy was 94.3% for random forest and 96.4% for support vector machines in the sex-mixed model, and 94.1% and 96.0%, respectively, in the four-way model. Support vector machines outperformed random forests in sex-specific classifications. Analysis of mean absolute Shapley Additive explanations (SHAP) values identified the subpubic angle and the height between the anterior superior iliac spine and the anteroinferior margin of the ischial tuberosity as the most influential variables. These findings demonstrate that CT-based pelvic morphometrics can reliably differentiate Indonesian and Japanese individuals and support their application in forensic identification when crania are unavailable.