Introduction:
Cephalometric analysis, a cornerstone of orthodontics and craniofacial surgery, traditionally involves manual radiograph tracing, a time-consuming and potentially variable process. Artificial intelligence (AI) offers a potential alternative for faster, more consistent analysis. This study compared AI-driven and manual cephalometric methods to assess agreement and identify discrepancies.

Material and methods:
This quantitative, comparative cross-sectional study was conducted in a private practice in Peshawar, Pakistan (August–November 2024), including 29 orthodontic patients who met specific criteria (good-quality cephalograms and absence of facial clefts/intra-oral appliances). Cephalometric radiographs were analyzed by two experienced dentists using manual tracing and by AI software (Audaxceph 6.0.50.3887). Five key angular measurements (SNA, SNB, ANB, FMA, and SN-Mp), used in Steiner’s and Tweed’s analyses, were compared. Inter-rater reliability for the manual tracings was assessed using intraclass correlation coefficients (ICCs).

Results:
Excellent inter-rater reliability was observed for manual tracings (ICCs > 0.90). Paired t-tests revealed no significant differences between manual and AI methods for SNA, SNB, ANB, and FMA. However, a statistically significant difference (p = 0.006) was found for SN-Mp.

Conclusions:
This study, comparing manual and AI-driven cephalometric analysis, found strong agreement for most key measurements (SNA, SNB, ANB, and FMA), suggesting AI’s potential to enhance clinical efficiency. The significant difference in SN-Mp, however, emphasizes the need for continued clinical oversight. A combined approach, integrating AI with clinical expertise, is recommended for optimal diagnostic accuracy and treatment planning.