Reliability and accuracy of Artificial intelligence-based software for cephalometric diagnosis. A diagnostic study

BMC Oral Health. 2024 Oct 28;24(1):1309. doi: 10.1186/s12903-024-05097-6.

Abstract

Background: Artificial intelligence (AI) is revolutionizing cephalometric diagnosis in orthodontics, streamlining the patient assessments. This study aimed to assess the reliability, accuracy, and time consumption of artificial intelligence (AI)-based software compared to a conventional digital cephalometric analysis method on 2D lateral cephalogram.

Methods: 408 lateral cephalometries were analysed using three methods: manual landmark localization, automatic localization, and semi-automatic localization with AI-based software. On each lateral cephalogram, 15 variables were selected, including skeletal, dental, and soft tissue measurements. The difference between the two AI-based software options (automatic and semi-automatic) was compared with the conventional digital technique. The time required to produce a complete cephalometric tracing was evaluated for each method using Student's t-test.

Results: Statistically significant differences in the accuracy of landmark positioning were detected among the three different techniques (p < 0,01). However, it is noteworthy that almost all of these differences were not clinically significant. There was a small difference in accuracy between the semi-automatic AI-based option and conventional digital techniques. Regarding the time used for each technique, the automatic version was the fastest, followed by the semi-automatic option and the conventional digital technique. (p < 0,000).

Conclusions: The study showed a statistical difference in accuracy between the conventional digital technique and two AI-based software alternatives, but these differences were not clinically significant except for specific measurements. The semi-automatic option was more accurate than the automatic one and faster than conventional tracing. Further research is needed to confirm AI's accuracy in cephalometric tracing.

Keywords: Artificial intelligence; Cephalometry; Software.

Publication types

  • Comparative Study

MeSH terms

  • Adolescent
  • Anatomic Landmarks / diagnostic imaging
  • Artificial Intelligence*
  • Cephalometry* / methods
  • Child
  • Female
  • Humans
  • Image Processing, Computer-Assisted / methods
  • Male
  • Reproducibility of Results
  • Software*