Accuracy and Reliability of Artificial Intelligence Cephalometric Analysis Software Compared to Manual Tracing

The Accuracy and Reliability of Orthodontic Cephalometry Analysis Using Web-Based Artificial Intelligence Program

This study compares the accuracy and reliability of artificial intelligence (AI) software for analyzing dental X-rays to the traditional manual tracing method used by dentists.

Lateral cephalometric radiographs are special X-rays of the head used in orthodontics (teeth straightening) to measure jawbone positions, tooth angles, and facial proportions. Traditionally, orthodontists manually trace these X-rays using pencil and paper to identify key landmarks and make measurements. This manual method is time-consuming and can vary between different practitioners or even when the same practitioner measures twice.

AI-based software can automatically identify these landmarks and perform measurements instantly. This study examined 40 dental X-rays to determine if the AI software (WeDoCeph) is as accurate and more reliable than manual tracing.

Each X-ray was measured twice - once manually by a trained examiner and once by AI software - at two different times (4 weeks apart). The researchers compared 15 different measurements, including 8 angles and 7 distances, to assess accuracy and reliability.

Study Overview

• Status: Completed
• Conditions: Orthodontic; Cephalometric Analysis; Cephalometry
• Intervention / Treatment: Diagnostic test: Manual Cephalometric Tracing; Diagnostic test: AI-Based Cephalometric Analysis (WeDoCeph Software)

Detailed Description

Lateral cephalometric analysis is essential for orthodontic diagnosis and treatment planning. The traditional manual tracing method involves identifying anatomical landmarks on radiographs using pencil, ruler, and protractor, which is subjective, time-consuming, and prone to intra- and inter-observer variability.

This diagnostic accuracy study evaluated the WeDoCeph AI-based cephalometric analysis software against conventional manual tracing. The study used a comparative repeated-measures design where each radiograph was analysed by both methods at two time points (T₀ and T₁, separated by 4 weeks) to assess both accuracy and reliability.

Sample size calculation was based on 95% power and a 0.05 significance level, resulting in 40 lateral cephalometric radiographs. All measurements included angular parameters (SNA, SNB, ANB, FMPA, MMPA, UIA, LIA, IIA) and linear parameters (A-N perpendicular, POG-N perpendicular, ANS-Me, SN, UFH, MxPI, MnPI).

Paired T-Test will be employed as the statistical analysis method for comparisons and Intraclass Correlation Coefficient (ICC) for reliability assessment. The study aimed to determine whether AI-based cephalometric analysis provides sufficient accuracy and superior reliability for clinical application in orthodontic practice.

• Study Type: Observational
• Enrollment (Actual): 40

Contacts and Locations

Study Locations

• Malaysia
  • Pahang
    • Kuantan, Pahang, Malaysia, 25200
      • Orthodontic Specialist Clinic, Kulliyyah of Dentistry

Participation Criteria

Eligibility Criteria

• Ages Eligible for Study: Child; Adult; Older Adult
• Accepts Healthy Volunteers: No

• Sampling Method: Non-Probability Sample

Study Population

The study population consisted of radiographs from orthodontic patients at various stages of treatment, including both pretreatment (initial diagnostic) and post-treatment radiographs. All radiographs were high-quality digital or digitized lateral cephalograms suitable for landmark identification and measurement. Patients with surgical rigid fixations, orthodontic appliances visible on radiographs, or dental prostheses were excluded to ensure clear visualization of anatomical landmarks. Additionally, radiographs of very poor quality or from patients with diagnosed syndromes or craniofacial deformities were excluded to maintain consistency in anatomical structure assessment.

The unit of analysis is the cephalometric radiograph rather than individual patients, as each radiograph represents a single diagnostic assessment.

Description

Inclusion Criteria:

• Pretreatment/post-treatment lateral cephalometric radiographs
• High-quality cephalograms with visible anatomical landmarks

Exclusion Criteria:

• Patients with surgical rigid fixations, orthodontic appliances and dental prostheses visible on radiographs
• Very poor quality/diagnostically unacceptable radiographs
• Patients with syndromes or with craniofacial deformities

Study Plan

How is the study designed?

Number of groups / cohorts

1

Cohorts and Interventions

Group / Cohort

Intervention / Treatment

Orthodontic Patients with Lateral Cephalometric Radiographs; Lateral cephalometric radiographs from 40 orthodontic patients collected between January 2023 and June 2023 from the Orthodontic Specialist Clinic. Each radiograph was analyzed using both manual tracing and AI-based software (WeDoCeph) at two time points (initial and 4 weeks later)

Diagnostic test: Manual Cephalometric Tracing; Conventional manual cephalometric analysis performed by trained examiner using traditional tracing technique. Lateral cephalometric radiographs are hand-traced in a darkened room using a view box for transillumination. A 25cm x 18cm radiographic film is used as the base, with a 21cm x 16cm matte acetate tracing paper taped over it. Hard and soft tissue cephalometric landmarks are manually identified and traced using a 0.3mm 2HB pencil. Angular measurements are obtained using a protractor, and linear measurements using a ruler. All 15 cephalometric measurements (8 angular: SNA, SNB, ANB, FMPA, MMPA, UIA, LIA, IIA; and 7 linear: A-N perpendicular, POG-N perpendicular, ANS-Me, SN, UFH, MxPI, MnPI) are calculated manually. Each radiograph is traced and analyzed twice at 4-week intervals by the same examiner to assess intra-examiner reliability.; Diagnostic test: AI-Based Cephalometric Analysis (WeDoCeph Software); Automated cephalometric analysis using WeDoCeph artificial intelligence-based software. Digital lateral cephalometric radiographs are imported as high-quality JPEG images into the software platform. The AI system automatically identifies and traces cephalometric landmarks using deep learning algorithms, then instantly generates all measurements based on the predefined parameters. The same 15 cephalometric measurements obtained in manual tracing (8 angular: SNA, SNB, ANB, FMPA, MMPA, UIA, LIA, IIA; and 7 linear: A-N perpendicular, POG-N perpendicular, ANS-Me, SN, UFH, MxPI, MnPI) are automatically calculated by the software. Each radiograph is analyzed twice at 4-week intervals using the previously uploaded digital images to assess reproducibility and consistency of the AI system. No manual landmark identification or measurement calculation is required.

What is the study measuring?

Primary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Intraclass Correlation Coefficient (ICC) for repeated manual measurements	ICC calculated for all 15 cephalometric measurements (8 angular and 7 linear) performed manually at two time points to assess intra-examiner reliability	Baseline (T₀) and 4 weeks later (T₁)
Intraclass Correlation Coefficient (ICC) for repeated AI measurements	ICC calculated for all 15 cephalometric measurements performed by WeDoCeph software at two time points to assess consistency	Baseline (T₀) and 4 weeks later (T₁)
Mean differences between manual and AI-based measurements at T₀	Paired T-Test comparison of all 15 measurements between manual tracing and AI analysis at initial time point	Baseline (T₀)
Mean differences between manual and AI-based measurements at T₁	Paired T-Test comparison of all 15 measurements between manual tracing and AI analysis at 4-week time point	4 weeks

Secondary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Angular Measurements	Comparison of angular cephalometric measurements between methods	Baseline (T₀) and 4 weeks (T₁)
Linear Measurements	Comparison of linear cephalometric measurements between methods	Baseline (T₀) and 4 weeks (T₁)
Inter-examiner Reliability	10% of radiographs were analyzed by three examiners to ensure inter-examiner agreement	During calibration phase

Collaborators and Investigators

• Sponsor: International Islamic University Malaysia

Publications and helpful links

General Publications

• Alqahtani H. Evaluation of an online website-based platform for cephalometric analysis. J Stomatol Oral Maxillofac Surg. 2020 Feb;121(1):53-57. doi: 10.1016/j.jormas.2019.04.017. Epub 2019 May 3.
• Kazimierczak W, Gawin G, Janiszewska-Olszowska J, Dyszkiewicz-Konwinska M, Nowicki P, Kazimierczak N, Serafin Z, Orhan K. Comparison of Three Commercially Available, AI-Driven Cephalometric Analysis Tools in Orthodontics. J Clin Med. 2024 Jun 26;13(13):3733. doi: 10.3390/jcm13133733.
• Lee JH, Yu HJ, Kim MJ, Kim JW, Choi J. Automated cephalometric landmark detection with confidence regions using Bayesian convolutional neural networks. BMC Oral Health. 2020 Oct 7;20(1):270. doi: 10.1186/s12903-020-01256-7.

Study record dates

Study Major Dates

• Study Start (Actual): January 2, 2023
• Primary Completion (Actual): June 30, 2023
• Study Completion (Actual): June 30, 2023

Study Registration Dates

• First Submitted: November 14, 2025
• First Submitted That Met QC Criteria: November 17, 2025
• First Posted (Actual): November 24, 2025

Study Record Updates

• Last Update Posted (Actual): November 24, 2025
• Last Update Submitted That Met QC Criteria: November 17, 2025
• Last Verified: January 1, 2023

More Information

Terms related to this study

• Other Study ID Numbers: IREC 2023-045; CHAIN 22-001-0001 (Other Grant/Funding Number: IIUM Kulliyyah of Dentistry Postgraduate CHAIN Grant)

Plan for Individual participant data (IPD)

• Plan to Share Individual Participant Data (IPD)?: NO
• IPD Plan Description: Individual participant data will not be made publicly available to protect patient privacy and confidentiality. The study involves radiographic images and associated measurements from orthodontic patients. Even with de-identification, radiographic images may be potentially identifiable. Data sharing was not included in the original ethics approval and informed consent process. Aggregate summary data and statistical results are available in the published manuscript.

Drug and device information, study documents

• Studies a U.S. FDA-regulated drug product: No
• Studies a U.S. FDA-regulated device product: No