AI-assisted semi-automated segmentation for tooth volume analysis in postmortem CT imaging: evaluation of forensic applicability.

Abstract

Objectives: This study investigates whether tooth volume measurements derived from postmortem computed tomography (PMCT) can provide discriminatory information relevant for forensic identification or demographic profiling. In particular, it evaluates whether tooth volume represents a potentially useful parameter independent of dental restorations.

Material and methods: 60 anonymized PMCT scans from the Zurich Institute of Forensic Medicine were analyzed. Of these, 39 scans were from males and 21 scans were from females (mean age; 37 years). Following strict inclusion and exclusion criteria, 1,254 untreated, fully developed teeth were segmented using a semi-automated workflow combining machine learning and manual correction. Tooth volumes were calculated and mixed models were applied to assess the influence of age and sex.

Results: Tooth volume varied substantially by tooth type, with posterior teeth showing larger volumes than anterior teeth. No evidence was found that tooth volume alone is sufficiently distinctive for direct identification. Tooth volume showed a small but significant negative correlation with age (p = 0.008 after Bonferroni adjustment) and was generally larger in men than in women (p < 0.001 for the differences between the sexes). However, these differences were not sufficient for reliable individual identification. Technical limitations due to metal restorations, artifacts and incomplete root development led to the exclusion of a considerable number of teeth.

Conclusions: Tooth volume derived from CT-based segmentations alone does not appear to be sufficiently discriminative for direct forensic identification. However, it may provide supportive information for demographic profiling when combined with additional morphological or segmentation-derived parameters.