Evaluation of an artificial intelligence–based algorithm for automated localization of craniofacial landmarks

• DOI: 10.1007/s00784-023-04978-4
• Publication Date: 2023-04-04T11:52:19Z
• Authors: Friederike Maria ...; Stephan Christian...; Stefan Raith; Tobias Pankert; Florian Peters; Michael Wolf; Frank Hölzle; Ali Modabber
• Keywords: Landmark; Ground truth

Due to advancing digitalisation, it is of interest develop standardised and reproducible fully automated analysis methods cranial structures in order reduce the workload diagnosis treatment planning generate objectifiable data. The aim this study was train evaluate an algorithm based on deep learning for detection craniofacial landmarks cone-beam computed tomography (CBCT) terms accuracy, speed, reproducibility.A total 931 CBCTs were used algorithm. To test algorithm, 35 located manually by three experts automatically 114 CBCTs. time distance between measured values ground truth previously determined orthodontist analyzed. Intraindividual variations manual localization using 50 analyzed twice.The results showed no statistically significant difference two measurement methods. Overall, with a mean error 2.73 mm, AI 2.12% better 95% faster than experts. In area bilateral structures, able achieve average.The achieved accuracy automatic landmark clinically acceptable range, comparable precision determination, requires less time.Further enlargement database continued development optimization may lead ubiquitous CBCT datasets future routine clinical practice.