Accuracy comparative study of automatic landmarking and diagnostic models on lateral cephalograms

Le résumé fourni par la source

BACKGROUND: The application of deep learning techniques in cephalometric analysis has become increasingly prominent. Although automatic landmarking models for cephalometric analysis have been developed, their accuracy still requires validation and relies heavily on clinicians to resolve discrepancies between results. To address these limitations, automatic diagnostic models have gained attention. However, there is no direct evidence establishing the superiority of one model over the other, especially the generalization and transferability. METHODS: Based on the same northern Chinese population external test dataset data and the data of the IEEE (Institute of Electrical and Electronics Engineers) 2015 ISBI (International Symposium on Biomedical Imaging) Grand Challenge dataset, we compared the performance, generalization ability, and transfer ability of the proposed two models, respectively. RESULTS: Our findings suggest that the automatic landmarking model outperforms the automatic diagnostic model in both external test dataset, with an accuracy of 90.80% on the IEEE dataset. CONCLUSIONS: In this study, the comparison was indirect, with each model having its strengths: the automatic landmarking model offers precise measurements, while the automatic diagnostic model provides quicker results. The choice between them depends on clinical needs, and future work should explore hybrid models to combine the advantages of both.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.