Original Research

Year : 2026 | Volume : 17 | Issue : 3 | Page : 27-40

Accuracy of a Locally Developed Automated Cephalometric Landmark-Detection System (Alphax) Compared With WebCeph: A Diagnostic-Accuracy Benchmarking Study Against Expert Manual Tracing

Hosam Zaza-1, Yehya Mostafa-2, Mostafa El-Dawlatly-3

1-Postgraduate researcher (PhD candidate), 2-Professor, Department of Orthodontics, Faculty of Oral and Dental Medicine, Future University in Egypt, New Cairo, Cairo, Egypt, 3-Associate Professor, Department of Orthodontics, Faculty of Dentistry, Cairo University, Giza, Egypt.

Address for Correspondence:

Dr. Hosam Zaza,

Department of Orthodontics,

Faculty of Oral and Dental Medicine,

Future University in Egypt, New Cairo, Cairo, Egypt.

E-Mail: hosam.zaazaa@fue.edu.eg

ABSTRACT

OBJECTIVE: Automated landmark detection is now widely used in cephalometric analysis, and several commercial platforms are available. A newly developed system must be tested against both an established platform and human experts before it can be used clinically. This study compared Alphax, a locally developed system that places cephalometric landmarks automatically on lateral cephalograms, with WebCeph (AssembleCircle, Seoul, South Korea), using expert manual tracing as the expert reference standard.

MATERIALS AND METHODS: Three hundred lateral cephalograms not used in development formed the test set; two calibrated orthodontists set the reference standard. Only the 13 landmarks that both systems place automatically (nine skeletal, four incisal) were analyzed; the eight further landmarks produced by Alphax were not assessed. Outcomes were per-landmark mean radial error (MRE) and successful-detection rate (SDR) at 2.0 to 4.0 mm. Because both systems were tested on the same radiographs, every comparison was paired. Systems were compared with the Wilcoxon signed-rank test, corrected for multiple comparisons; detection rates were compared with McNemar's test.

RESULTS: Expert intra- and inter-observer MRE were 0.42 and 0.96 mm. Alphax achieved an overall MRE of 0.97 mm and a 2-mm SDR of 93.4%, equivalent to the expert reference (mean difference 0.01 mm). WebCeph was less accurate (MRE 1.31 mm, SDR 82.2%), and Alphax was significantly more accurate at all 13 landmarks (all adjusted P < 0.05). All seven derived cephalometric angles were closer to expert values with Alphax.

CONCLUSION: Across the 13 landmarks common to both systems, Alphax was significantly more accurate than WebCeph at every landmark, and tracked expert manual tracing closely, matching the two-observer consensus and falling 0.13 mm short of a single observer. Automated tracing was adequate for screening, but landmark verification remained advisable at the most error-prone points.
KEY-WORDS: Artificial intelligence; Automated landmark detection; Cephalometry; Reproducibility of results; Orthodontic diagnosis; WebCeph.

SOURCES OF FUNDING

This investigation received no external funding, grant, equipment or drug support from any public, commercial or not-for-profit agency.

CONFLICT OF INTEREST

The authors report no commercial, proprietary or financial interest in any of the products or companies described in this article. The authors hold no equity in, and receive no royalties from, any commercial cephalometric platform, including WebCeph (AssembleCircle, Seoul, South Korea). No author has any conflict of interest with products that compete with those mentioned in the manuscript.

How to cite this article: Hosam Zaza, Yehya Mostafa, Mostafa El-Dawlatly. Accuracy of a Locally Developed Automated Cephalometric Landmark-Detection System (Alphax) Compared With WebCeph: A Diagnostic-Accuracy Benchmarking Study Against Expert Manual Tracing.

Int J OrthodRehabil 2026; 17 (3): 27-40.

Doi: 10.56501/Intjorthodrehabil.172.ijor0133