Assessment of AI-based cephalometric landmark recognition on lateral cephalometric radiographs
Source: PubMed Central Open Access, NCBI / U.S. National Library of Medicine
Background Artificial intelligence (AI) strategies have been proposed for automatic landmark recognition applications with the expectation of simplifying and improving cephalometric analysis through accurate and consistent landmarks identification. The aim of the current study was to evaluate the accuracy and reliability of AI in identifying cephalometric landmarks for orthodontics cephalometric tracing. Materials and methods A total of 506 lateral cephalometric radiographs of patients aged 14–40 years were retrieved from the archives of Dubai Dental Hospital from the period of 2018–2021. Radiographs were traced using Dolphin imaging software, and the AI program (AI Ceph) was trained to identify 19 key cephalometric landmarks. Manual tracings were compared with AI-generated ones to evaluate the accuracy and reliability of the AI Ceph program. Results High agreement was observed between manual and AI-generated tracings. The highest concordance was detected for Menton (96.6%), whereas the lowest was observed for Nasion (86.6%). The deviation between the two methods was within 2 mm for all landmarks except Glabella (< 0.02). Sensitivity values exceeded 80% for all evaluated landmarks except for Nasion and A-point where they showed a sensitivity of 72.7% and 78.8% respectively, demonstrating acceptable accuracy and reliability of the AI system. Conclusions The tested AI method was accurate and reliable for identifying cephalometric landmarks, with a clinically acceptable 2 mm dev
Abstract
Background Artificial intelligence (AI) strategies have been proposed for automatic landmark recognition applications with the expectation of simplifying and improving cephalometric analysis through accurate and consistent landmarks identification. The aim of the current study was to evaluate the accuracy and reliability of AI in identifying cephalometric landmarks for orthodontics cephalometric tracing. Materials and methods A total of 506 lateral cephalometric radiographs of patients aged 14–40 years were retrieved from the archives of Dubai Dental Hospital from the period of 2018–2021. Radiographs were traced using Dolphin imaging software, and the AI program (AI Ceph) was trained to identify 19 key cephalometric landmarks. Manual tracings were compared with AI-generated ones to evaluate the accuracy and reliability of the AI Ceph program. Results High agreement was observed between manual and AI-generated tracings. The highest concordance was detected for Menton (96.6%), whereas the lowest was observed for Nasion (86.6%). The deviation between the two methods was within 2 mm for all landmarks except Glabella (< 0.02). Sensitivity values exceeded 80% for all evaluated landmarks except for Nasion and A-point where they showed a sensitivity of 72.7% and 78.8% respectively, demonstrating acceptable accuracy and reliability of the AI system. Conclusions The tested AI method was accurate and reliable for identifying cephalometric landmarks, with a clinically acceptable 2 mm deviation. AI-assisted cephalometric tracing may serve as a useful adjunctive tool to support orthodontists in diagnosis and treatment planning.