AI-automated Cephalometric Tracing: A New Normal in Orthodontics?

Cephalometric analysis has typically been used to evaluate lateral skull radiographs taken with a cephalostat to determine the skeletal pattern and assess treatment difficulties in orthodontics. Nowadays, Cone beam computed tomography (CBCT) data can be used for 3D cephalometrics. This study compares the performance of AI-automated cephalometric tracing with human-operated digital tracing using the Invivo 7.1.2 software on a sample of 120 scans evaluated by 5 observers (4 human and AI) for 14 cephalometric variables. Measurements were repeated for paired tests. ANOVA was used to minimize type I error. Normality of the data was tested using the Shapiro-Wilk test. The Wilcoxon test was used to compare measurement times between humans and AI. This study found that there were statistically significant differences between human observers and AI in 6 of 14 variables. The AI measurements were higher than the human measurements for these variables. The duration of measurements was also statistically significantly shorter for human observers than for AI. Now at the dawn of an era of AI-assisted diagnostics, we temporarily experience translational period of AI-automated algorithms being less accurate and slower than humans, albeit they will likely become the new normal in orthodontic treatment planning. This study also highlights the importance of rigorous scientific evaluation by independent researchers of incoming AI-automated cephalometric tracing solutions before they are considered clinically appropriate.