All study methods received approval from the local ethics committee of Jena University Hospital (registration number 2019-1505-MV) and adhered to relevant guidelines. Funding was provided by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) – 514572362.

The retrospective study analyzed 819 cranial computed tomography (CCT) examinations (age range 10-99 years, 321 female, 452 male and 46 unknown) of 722 individuals acquired between 11/2016 and 05/2023. Six regions were defined (refer to Figure 1), and their CT slices were manually selected from the CT series: 

1. lower row of teeth,
2. upper row of teeth,
3. end of maxilla,
4. cervical spine,
5. maximal representation of maxillary sinus and
6. maximal representation of eye structures
7. OPGs (reference)

Fig 1: CT single slices (regions) used in the study: A) lower row of teeth, B) upper...

Large metal artifacts, such as those caused by dental implants, or other artifacts resulting from factors like movement during the examination, for example, led to the inability to definitively locate all of the aforementioned categories. Consequently, these specific images could not be exported from the CT series for the subsequent identification process. For comparative analysis, an additional 1,771 OPGs acquired from the same individuals between 12/2000 and 05/2023 were included.

The image processing and identification methods were adopted from previous studies on OPGs [1; 2]. Unlike OPGs, no cropping of the edges of the CT images was performed [1]. The algorithm AKAZE was applied to extract CV features. The image processing and evaluation has been performed on a standard computer (Intel® i7, 4 x 3.1 GHz CPU, 16 GB LPDDR3 2133 MHz RAM).

For unique person identification, between 50 and 69 CT slices per region (the most recent ones from individuals with at least two examinations) were compared with up to 818 database entries (A: 697, B: 651, C: 818, D: 805, E: 814, F: 814). Additionally, 410 OPGs were matched with 1759 OPGs from the same individuals. A concordance metric for CV feature matching was calculated: 

score = (matching points / number of keypoints) [%]

The number of keypoints corresponds to the maximum number of CV features in the postmortem image and results in a normalization of the score between 0 and 100%.