Left-Right Consistency and Quantification of Craniofacial Asymmetry by Two Fully Automatic AI-Powered...

Area of Science:

• Dentistry
• Medical Imaging
• Artificial Intelligence

Background:

• Cone beam computed tomography (CBCT) enables automatic 3D cephalometry in orthodontics and orthognathic surgery.
• The reliability of bilateral measurements and asymmetry quantification by AI systems requires systematic evaluation.

Purpose of the Study:

• To assess the side-specific agreement of two commercial AI cephalometry systems against manual 3D tracing.
• To evaluate the fidelity of AI-based asymmetry quantification compared to manual analysis.

Main Methods:

• Thirty-one CBCT scans were analyzed using manual 3D tracing (InVivo), an integrated AI module (AI InVivo), and a cloud-based system (Diagnocat).
• Side-specific agreement was quantified using intraclass correlation coefficient (ICC), Bland-Altman analysis, and equivalence tests.
• Asymmetry fidelity was assessed by ICC of the right-left difference (Δ = R - L) against manual Δ, Pearson correlation, and sensitivity/specificity.

Main Results:

• AI systems applied systematic errors symmetrically, resulting in equal per-side accuracy.
• Mandibular plane angle showed excellent ICC (0.95-0.97) and equivalence for both systems.
• Gonial angle equivalence was achieved with AI InVivo (ICC 0.92-0.93) but not Diagnocat (ICC 0.40).
• Asymmetry agreement was generally weaker than per-side accuracy; both systems under-detected clinically relevant gonial asymmetry.

Conclusions:

• Fully automatic AI cephalometry ensures symmetrical systematic errors, maintaining equal accuracy for right and left measurements.
• Asymmetry quantification fidelity is a separate issue, proving unreliable for landmark-sensitive parameters like the gonial angle.
• Clinician verification is crucial before utilizing automatic AI analysis for asymmetry assessment.