Accuracy of a real-time registration system based on orthodontic landmarks

Huang Juneng; Zeng Jingjie; Peng Ying; Lu Wenrong; Tang Min; Zhang Xuejun

doi:10.12307/2026.624

Abstract：BACKGROUND:The superimposition of consecutive cephalometric lateral radiographs is commonly used in orthodontics to assess the outcomes of orthodontic/orthopedic treatments and changes in growth and development.The key factors affecting the accuracy of cephalometric superimposition are landmark localization errors and registration methods.Currently,there is no highly accurate registration system. OBJECTIVE:To study a real-time registration system that guides the adjustment of landmark localization on cephalometric lateral radiographs to assist in improving registration accuracy and evaluate the precision of this registration method. METHODS:Pre-and post-treatment cephalometric lateral radiographs of 40 adult patients were selected according to the inclusion criteria.Landmark localization on the pre-and post-treatment radiographs was performed using Labelme software.Real-time visualized registration software was developed using Python,Opencv,and PyQT.Registration was performed using the SellaNasion(SN)superimposition method,the Least Squared Averaged 5 Landmarks(LS-5)method,and the Real-time Least Squared Averaged 5 Landmarks(LS-5R)method.The distance errors and the success detection rate of landmarks anterior nasal spine(ANS),posterior nasal spine(PNS),upper alveolar seat point(A),lower alveolar seat point(B),point of anterior chin(Pog)were measured and compared to evaluate the registration accuracy. RESULTS AND CONCLUSION:The LS-5R method exhibited the smallest landmark distance errors.Compared with the SN method,all landmark distance errors showed statistically significant differences(P<0.05).The distance errors of PNS,A,B,and Pog landmarks with the LS-5R method were statistically different from those with the LS-5 method(P<0.05),and the differences increased in sequence.The success detection rate of automatic landmark detection within 1mm,2mm,and 3mm accuracy ranges was higher for the LS-5R method than for the SN superimposition method.The LS-5R method has smaller superimposition errors compared with the traditional SN method,making it a potentially more reliable method for superimposing consecutive cephalometric lateral radiographs in clinical practice.