Comparison of Digital Analysis and Artificial Intelligence for Cephalometric Tracing

This prospective observational study aims to evaluate the accuracy, reproducibility, and clinical reliability of artificial intelligence (AI)-based cephalometric analysis systems compared with digital manual tracing. Patients whose lateral cephalometric radiographs were previously acquired for orthodontic diagnostic purposes at the Unit of Orthodontics and Paediatric Dentistry, University of Pavia, will be retrospectively selected according to predefined inclusion and exclusion criteria. Written informed consent for the use of clinical records for research purposes will be obtained from all participants or their legal guardians. A total of 100 standardized digital lateral cephalometric radiographs will be included in the study. Each radiograph will be analysed using Delta-Dent software with manual landmark identification and three fully automated AI-based software systems: WebCeph™, QuantX, and Smartee. Cephalometric analyses will be performed without manual adjustment of landmarks in the AI-based systems. Ten cephalometric parameters representative of sagittal, vertical, dental, and soft tissue relationships will be evaluated, including SNA, SNB, ANB, SN-GoGn, L1-GoGn, U1-ANSPNS, nasolabial angle, facial angle, Wits appraisal, and N-Me. Manual digital tracing performed by a single experienced orthodontist will be considered the reference method. Intra-rater reliability will be assessed using intraclass correlation coefficient (ICC). Statistical analysis will be conducted using R software (version 3.1.3; R Foundation for Statistical Computing, Wien, Austria). Descriptive statistics will be calculated for all variables. Normality of data distribution will be assessed using the Kolmogorov-Smirnov test. Comparisons among the different methods will be performed using the Friedman test followed by Dunn's post hoc test. Statistical significance will be predetermined at p < 0.05.

• Cephalometric Analysis (Diagnostic test): All included lateral cephalometric radiographs will undergo cephalometric analysis using both digital manual tracing and artificial intelligence-based automated systems. Manual digital tracing will be performed with Delta-Dent software by a single experienced orthodontist through manual identification of cephalometric landmarks. The same radiographs will subsequently be analysed using three fully automated AI-based software programs (WebCeph™, QuantX, and Smartee) without manual correction of landmark positioning. No therapeutic intervention or modification of patient treatment will be performed, as this is an observational comparative study based exclusively on retrospective analysis of radiographic records.

Primary outcomes

• Agreement between AI-based cephalometric analysis and digital manual tracing [Baseline]: Agreement between cephalometric measurements obtained with AI-based software systems and digital manual tracing will be assessed using the intraclass correlation coefficient (ICC) and differences in angular and linear cephalometric measurements.

Secondary outcomes

• SNA [Baseline]
• SNB [Baseline]
• ANB [Baseline]
• SN-GoGn [Baseline]
• L1-GoGn [Baseline]
• U1-ANSPNS [Baseline]
• Nasolabial angle [Baseline]
• Facial angle [Baseline]
• Wits appraisal [Baseline]
• N-Me [Baseline]

Sex: All

Inclusion Criteria:

* Availability of digital lateral cephalometric radiographs of adequate diagnostic quality
* Radiographs acquired with patients in centric occlusion and proper head positioning using a cephalostat
* Patients of any age and sex
* Absence of congenital or acquired craniofacial anomalies
* No previous orthodontic treatment
* No previous orthognathic surgical treatment
* Absence of agenesis of incisors or first molars
* Absence of supernumerary teeth overlapping the region of interest

Exclusion Criteria:

* Radiographs presenting artifacts or inadequate visualization of anatomical structures
* History of significant craniofacial trauma
* Radiographs acquired without a cephalostat
* Presence of severe skeletal asymmetries
* Incomplete clinical or radiographic records
* Radiographs unsuitable for manual or AI-based cephalometric landmark identification