Evaluation of cephalometric landmarks using artificial intelligence compared to the manual method

There is growing interest in cephalometric methods based on artificial intelligence (AI), as they can significantly reduce clinicians’ workload and increase productivity. However, the literature indicates that further research is required to bridge the gap between operator-performed and AI-performed cephalometry. The aim of this study was to compare and evaluate the reproducibility and accuracy of three cephalometric methods: Manual, WebCeph™ and AngelAligner™. Thirty cephalograms from 30 patients (50% male, 50% female) were analyzed. Thirty-five landmarks were recorded using three different methodologies (Manual, WebCeph™, and AngelAligner™) at two time points (T1 and T2). The method with the highest number of cephalometric landmarks showing an excellent successful detection rate (SDR) was the manual method, in which 18 of the 35 points demonstrated an SDR < 1 mm in 80% of the analyzed cephalograms, thus being the most accurate. However, greater T1–T2 variations were observed with the manual method. The AI methods showed high and statistically significant correlations for all landmarks (p < 0.01), except for Pg’. The manual method showed low and non-significant correlations for all points except Na’ (p < 0.05) and UL (p < 0.01). Fully automated, AI-based cephalometric methods are more reproducible than manual methods. In terms of accuracy, certain AI-based point detections may be comparable to those of an expert, although supervision remains necessary to ensure precise and reliable results.