CT-Based Segmentation as a Basis for Individual Structural Comparisons for Person Identification

3D Structural Analysis for Person Identification

Abstract

The analysis of complex anatomical structures based on three-dimensional imaging data represents a central task in medical image processing. In particular, structures with high interindividual variability require robust segmentation and modeling approaches to enable reproducible morphometric analyses. The sphenoid sinus occupies a special position in this context, as its deep location at the skull base largely protects it from external influences while simultaneously exhibiting pronounced individual variability.

In this study, a comprehensive volumetric workflow for segmentation, three-dimensional reconstruction, and geometric analysis of the sphenoid sinus based on cranial CT data is presented. The methodological focus lies on standardized preprocessing, semi-automatic segmentation, and deep-learning-based volumetric segmentation using the nnU-Net framework.

The results demonstrate that the combination of intensity-based normalization, geometric standardization, and data-adaptive segmentation enables the generation of consistent and comparable 3D models. Stable segmentation performance was achieved even under markedly heterogeneous acquisition conditions. The proposed approach therefore provides a robust methodological foundation for further automated morphometric comparison procedures.