Gupta Shantam, Shetty Shravan, Natarajan Srikant, Nambiar Supriya, Mv Ashith, Agarwal Saloni
Department of Orthodontics and Dentofacial Orthopaedics, Manipal College of Dental Sciences Mangalore, Manipal Academy of Higher Education, Manipal, Karnataka, India.
Department of Oral Pathology and Microbiology, Manipal College of Dental Sciences Mangalore, Manipal Academy of Higher Education, Manipal, Karnataka, India.
J Clin Exp Dent. 2024 Jan 1;16(1):e11-e17. doi: 10.4317/jced.60899. eCollection 2024 Jan.
This study compared the accuracy and speed of cephalometric analysis using an artificial intelligence web-based method and a smartphone app-based system with manual cephalometric analysis as the reference standard.
In this cross-sectional study, the lateral cephalograms were analysed using four methods: manual tracing, smartphone app tracing, artificial intelligence web-based automated tracing without manual landmark identification correction and artificial intelligence web-based automated tracing with manual landmark identification correction. The principal investigator obtained linear and angular cephalometric measurements to compare the accuracies of the four methods being assessed. Additionally, the duration required for landmark identification and subsequent analysis was recorded.
The analyses included 40 lateral cephalograms that were selected based on the inclusion and exclusion criteria. Very good to excellent agreement was observed in the accuracies of the artificial intelligence web-based and smartphone app-based systems compared with manual tracing (interclass correlation coefficient values ranging from 0.707 to 0.9, < 0.001). Of the artificial intelligence web-based systems, the method without correction of automated landmark detection showed less reliable measurements than the other methods. Cephalometric analysis using artificial intelligence web-based and smartphone app-based systems consumed less time than manual tracing (< 0.001).
Artificial intelligence web-based automated tracing with manual landmark identification correction and smartphone-based app provide results that are comparable to those from the manual tracing method. However, artificial intelligence web-based systems require improvements in terms of automated landmark identification to obtain results that are similar to those from the other methods being assessed. Artificial Intelligence, Cephalometry, Computer software, Mobile application.
本研究将基于人工智能网络的方法和基于智能手机应用程序的系统进行头影测量分析的准确性和速度与以手工头影测量分析作为参考标准进行了比较。
在这项横断面研究中，使用四种方法对头侧位X线片进行分析：手工描记、智能手机应用程序描记、基于人工智能网络的自动描记（不进行手动标志点识别校正）以及基于人工智能网络的自动描记（进行手动标志点识别校正）。主要研究者获取线性和角度头影测量值，以比较所评估的四种方法的准确性。此外，记录标志点识别及后续分析所需的时间。
分析纳入了根据纳入和排除标准选择的40张头侧位X线片。与手工描记相比，基于人工智能网络的系统和基于智能手机应用程序的系统在准确性方面观察到非常好至极好的一致性（组内相关系数值范围为0.707至0.9，P<0.001）。在基于人工智能网络的系统中，未进行自动标志点检测校正的方法所显示的测量结果比其他方法的可靠性更低。使用基于人工智能网络的系统和基于智能手机应用程序的系统进行头影测量分析比手工描记耗时更少（P<0.001）。
基于人工智能网络的自动描记并进行手动标志点识别校正以及基于智能手机的应用程序所提供的结果与手工描记方法的结果相当。然而，基于人工智能网络的系统在自动标志点识别方面需要改进，以获得与所评估的其他方法相似的结果。人工智能、头影测量、计算机软件、移动应用程序。
