Akademik Veri Yönetim Sistemi
International Congress of DentoMaxilloFacial Radiology, London, İngiltere, 24 Haziran 2025, ss.16, (Özet Bildiri)
Abstract:
Materials and Methods: A total of 195 CBCT volumes containing the dento-alveolar region were obtained from the radiological archive of Eskisehir Osmangazi University Faculty of Dentistry in September 2024. Pulp anatomies showing partial obliteration were included to reflect clinical reality. Images with severe metal and motion artifacts that could affect the labeling process were excluded, resulting in 145 CBCT volumes used for labeling and training. Annotations were performed using the CranioCatch tool (Eskisehir, Turkiye). The dataset was divided into training (80%), validation (10%), and test (10%) subsets. The nnU-Netv2 segmentation model was trained for 1,000 epochs.
Results: The nnU-Netv2 model demonstrated reliable performance in detecting and segmenting pulp in CBCT images. A total of 11 CBCT volumes were used to test the model’s performance. It achieved a precision of 0.71, recall of 0.75, and a Dice score of 0.71. The model effectively identified pulp structures across various anatomical presentations, including partially obliterated canals, supporting its potential utility in clinical diagnostics.
Conclusion: The nnU-Netv2-based AI model automates the detection and segmentation of dental pulp in CBCT images with reasonable accuracy, enhancing diagnostic consistency and aiding in treatment planning. The inclusion of partially obliterated pulp anatomies in the labeling process highlights the model's relevance to clinical scenarios. Future research incorporating larger datasets and additional anatomical features could refine the model’s capabilities and broaden its clinical relevance.