Table 2.

Characteristics of the studies.

Authors/Years	Type of Study	Type of AI	Materials and Methods	Results
Galina Bulatova et al., 2021 [37]	Retrospective study	AI software Ceppro DDH Inc. (Seoul, Korea)	Lateral cephalograms were analyzed by a calibrated senior orthodontic resident using Dolphin Imaging® and the same images were uploaded to the AI software Ceppro DDH.	There was no statistical difference in manually analyzed CLs and those obtained by AI.
Young Hyun Kim et al., 2021 [38]	Retrospective study	The developed DL model has a two-step structure.	Two examiners manually identified the 13 most important CLs to set as references. The landmarks were automatically measured using the proposed model in lateral cephalometric images.	The proposed DL model can perform fully automatic identification of CLs.
Thaísa Pinheiro Silva et al., 2022 [39]	Retrospective study	CEFBOT (RadioMemory Ltd., Belo Horizonte, Brazil)	An expert and CEFBOT evaluated the 66 landmarks and 10 linear and angular measures featured in Arnett’s analysis on the radiograph.	CEFBOT (https://www.radiomemoryglobal.com/#h.r8d6r24868b accessed on 14 November 2023) software can be considered a promising tool.
Felix Kunz et al., 2020 [5]	Retrospective study	A customized open-source CNN DL algorithm (Keras and Google TensorFlow) is directed toward analyzing visual imagery and has an input layer, multiple hidden layers, and an output layer.	Both AI and each examiner analyzed 12 orthodontic parameters based on cephalometric images.	No clinically relevant difference was noticed between the two analyses.
Jaerong Kim et al., 2021 [40]	Retrospective study	A cascade network consisting of ROI detection and landmark prediction.	Two orthodontists evaluated 100 lateral cephalograms and the mean of these values was considered the gold standard. The DL model evaluated 3150 lateral cephalograms.	The overall automated detection error was 1.36 ± 0.98. The accuracy of CL recognition was comparable with that made by two orthodontists with more than 10 years of clinical experience.
Sangmin Jeon et al., 2021 [41]	Retrospective study	CephX for the AI analysis.	The cephalograms were analyzed with V-ceph for the conventional CA and with CephX for the AI analysis.	Variations were found in saddle angle, linear measurements of maxillary incisor to NA line, and mandibular incisor to NB line.
Mehmet Uğurlu et al., 2022 [42]	Retrospective study	AI system (CranioCatch, Eskisehir, Turkey).	A CNN-based AI algorithm for automatic CL detection was developed and used to detect CLs.Then, an orthodontist with 9 years of experience analyzed the CA of the AI.	There were no statistical differences between manual identification and AI groups in 11 out of 16 points. AI increased the efficiency of CL identification.
Gökhan Çoban et al., 2022 [43]	Retrospective study	WebCeph was used for AI-based CA.	Differences between using the semi-automated software Dolphin® (v. 11.5, Chatsworth, CA, USA) and WebCeph (WEBCEPH™, Artificial Intelligence Orthodontic & Orthognathic Cloud Platform, South Korea, 2020) software for each CL.	It was determined that there was a noticeable change between SNB, ANB, and SN.PP, U1.SN, U1-NA, U1.NA, L1-APog, IMPA, L1-NB, and ULE.
Ioannis A Tsolakis et al. [44]	Retrospective study	CS imaging V8 software was used for AI-based CA.	The difference between using semi-automated software Dolphin® 3D Imaging program (version 11.0) and CS imaging V8 software for each CL.	There were no significant differences between the two methods (p > 0.0027) for the SN-MP, U1-SN, SNA, SNB, ANB, L1-NB, SNPg, ANPg, SN/ANS-PNS, SN/GoGn, U1/ANS-PNS, L1-APg, U1-NA, and L1-GoGn landmarks.
Britta Ristau et al., 2022 [46]	Retrospective study	AudaxCeph®’s automatic tracing software.	The difference between AudaxCeph®’s automatic tracing and a semi-automated approach by human examiners using the same software.	AudaxCeph® was a reliable resource for clinicians in analyzing orthodontic cases, even if there were unreliable points, such as Porion, Orbitale, U1 apex, and L1 apex.
Mostafa El-Dawlatly et al., 2023 [47]	Retrospective study	WebCeph software and OnyxCeph software.	Lateral cephalometric radiographs were evaluated.	Fewer differences were obtained with the modified WebCeph software method than with the OnyxCeph method.
Pamir Meriç et al., 2020 [49]	Retrospective study	Dolphin Imaging® 13.01, app-aided tracing using the CephNinja 3.51 app, and fully automated web-based tracing with CephX.	Three methods were used to execute cephalometric measurements: Dolphin Imaging® 13.01, app-aided tracing using the CephNinja 3.51 app, and fully automated web-based tracing with CephX.	Manual correction of CephX landmarks gave similar outcomes to digital tracings using CephNinja and Dolphin®.