Abstract Purpose Lateral photographs are routinely evaluated as part of orthodontic diagnostics and treatment planning. Anthropometric measurements based on soft tissue landmarks are taken to evaluate facial features and attractiveness. The detection of these landmarks is a process performed by specialists and prone to intra- and inter-annotator variability. The aims of this investigation were (1) to train an artificial intelligence (AI) algorithm to automatically detect landmarks on lateral photographs, (2) to establish a high-quality gold standard dataset to evaluate landmark detection accuracy, and (3) to compare the performance of AI with that of clinical experts. Methods The AI algorithm was trained on a dataset of 991 photographs, with three clinical experts annotating 14 soft tissue landmarks on each photograph. Eleven experts annotated a separate dataset of 56 photographs, to establish the gold standard. Metric scaling of the photographs was achieved by transferring scaling from corresponding lateral cephalograms. Based on the detected landmarks, 11 anthropometric measurements were taken, and the performance of the experts and AI was compared against the gold standard by comparing errors from the ground truth using Mann–Whitney U tests. Results At a 2.0 mm threshold, the AI model achieved successful detection rates exceeding 95% for 12 of 14 landmarks. Compared with individual expert annotations, AI predictions showed reduced variability and lower mean radial errors for landmarks with high inter-annotator disagreement. Anthropometric measurements derived from AI predictions demonstrated smaller absolute errors than expert-based measurements. Conclusion This study demonstrates that AI-based landmark detection on lateral photographs can achieve accuracy comparable to expert annotations, demonstrating greater consistency for those landmarks exhibiting high inter-annotator variability.
Nordblom et al. (Fri,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: