Assessing AudaxCeph®’s cephalometric tracing technology versus a semi-automated approach for analyzing severe Class II and Class III skeletons

Objective

To evaluate the accuracy and precision of the AudaxCeph® fully automated software in identifying cephalometric landmarks on lateral cephalograms of Class II and Class III skeletal relationships, comparing its performance against experienced orthodontists using manual tracing within the same software environment.

Material and methods

Sixty cephalograms depicting severe Class II or Class III skeletal discrepancies were assessed by two board-certified orthodontists and AudaxCeph®’s artificial intelligence automatic tracing software. Among these, 40 cases were classified as Class II and 20 as Class III. An X–Y axis was established at the bottom left corner of each cephalogram, and subsequent X and Y coordinates for the landmarks were exported to Excel. Thirteen cephalometric landmarks were identified and used for comparing manual and automatic tracing methods, with no alteration of landmark positions post-tracing. Measures of the X coordinate, Y coordinate, and radial distance for each landmark were compared using t-tests for equivalence with a 2 mm margin, both against AudaxCeph®’s positions and intra-operator reliability.

Results

Analysis revealed that while most operator measurements closely approximated AudaxCeph® values, discrepancies exceeding 2 mm were notable at Gonion and Porion landmarks. Slight variability was noted in one instance during intra-examiner evaluation at the Gonion landmark.

Conclusions

This study concludes that AudaxCeph®’s artificial intelligence-driven automatic tracing of cephalograms offers a reliable and accurate method for orthodontic treatment planning across various skeletal types and severities. On average, it exhibits minimal discrepancies exceeding 2 mm compared to manual operators, with notable variations observed primarily at the Gonion and Porion landmarks. While AudaxCeph® is an acceptable tool for cephalometric landmark location, it's accuracy still require the practitioner to verify some less reliable landmark locations.