Background: Human identification in forensic contexts becomes challenging when fingerprints, dental records, or DNA are unavailable. The sphenoid sinus, owing to its protected anatomical location and high inter-individual variability, offers potential as a biometric marker. Methods: We first replicated a YOLOv8-nano segmentation experiment using annotated CT scans in DICOM data format. Images were preprocessed with soft-tissue windowing and converted into high-quality PNGs for training. Model performance was evaluated with standard segmentation metrics. Based on observed limitations, we decided for 3D modelling approaches involving manual, semi-automatic, and normalization-based segmentation pipelines. Results: The YOLOv8 model achieved high validation performance (mAP50 = 0.921; true positive rate = 91%; true negative rate = 100%), which is promising for the future implementation. However, its reliability for consistent sphenoid sinus segmentation was limited by anatomical complexity and heterogeneous datasets. In contrast, 3D modelling methods produced more robust and accurate reconstructions of the sphenoid sinus. Conclusion: While deep learning-based 2D segmentation provides a strong baseline for sinus analysis, 3D modelling approaches currently offer greater reliability for forensic applications. Combining both strategies may pave the way toward fully automated, scalable identification frameworks based on sphenoid sinus morphology.

Reliable personal identification remains a major challenge in forensic medicine, particularly in cases involving decomposition, thermal injury, or absence of DNA and dental records. In this context, anatomically protected and morphologically unique structures such as the sphenoid sinus may serve as valuable biometric markers. The aim of this study was to evaluate the forensic applicability of sphenoid sinus morphology using computed tomography (CT), automated segmentation, and three-dimensional (3D) computational analysis. The proposed framework integrates CT-based image preprocessing, automated 3D segmentation using nnU-Net architecture, and geometric comparison of reconstructed sphenoid sinus models through point cloud analysis and deep learning approaches. Morphological variability, spatial configuration, and structural stability of the sphenoid sinus were analyzed as discriminative biometric features. The GSA-Net–based identification model demonstrated high recognition performance, achieving Top-1 accuracies exceeding 97.8% and Top-3 accuracies up to 100% in controlled datasets. The results support the concept that the sphenoid sinus possesses sufficient individuality and anatomical preservation to enable reliable ante-mortem and post-mortem identification. The study highlights the potential of integrating automated segmentation and AI-driven 3D analysis into forensic workflows and emphasizes the importance of standardized imaging protocols and larger annotated datasets for future clinical and forensic implementation.

Die Auswertung komplexer anatomischer Strukturen auf Basis dreidimensionaler Bilddaten stellt eine zentrale Aufgabe der medizinisch-technischen Bildverarbeitung dar. Insbesondere Strukturen mit hoher interindividueller Variabilität erfordern robuste Segmentierungs- und Modellierungsverfahren, um reproduzierbare morphologische Analysen zu ermöglichen. Der Sinus sphenoidalis nimmt hierbei eine besondere Stellung ein, da er aufgrund seiner tiefen Lage an der Schädelbasis weitgehend vor äußeren Einflüssen geschützt ist und zugleich ausgeprägte individuelle Unterschiede aufweist. In dieser Arbeit wird ein durchgängiger volumetrischer Workflow zur Segmentierung, dreidimensionalen Rekonstruktion und geometrischen Analyse des Sinus sphenoidalis auf Basis kranialer CT-Daten vorgestellt. Der methodische Schwerpunkt liegt auf standardisierter Vorverarbeitung, semi-automatischer Segmentierung sowie Deep-Learning-basierter Volumensegmentierung unter Einsatz des nnU-Net-Frameworks. Die Ergebnisse zeigen, dass durch eine Kombination aus intensitätsbasierter Normalisierung, geometrischer Vereinheitlichung und datenadaptiver Segmentierung konsistente und vergleichbare 3D-Modelle erzeugt werden können. Auch bei deutlich heterogenen Akquisitionsparametern ließ sich eine stabile Segmentierungsleistung erzielen. Der vorgestellte Ansatz bietet damit eine belastbare Grundlage für weiterführende automatisierte morphologische Vergleichsverfahren.