AI-Assisted 3D diagnosis of impacted maxillary canines: A validation study.
Authors
Affiliations (12)
Affiliations (12)
- Department of Orthodontics and Pediatric Dentistry, School of Dentistry, University of Michigan, Ann Arbor, MI, USA.
- Department of Orthodontics, Bauru Dental School University of São Paulo, Alameda Dr. Octávio Pinheiro Brisolla, 9-75 ZIP Code, Bauru, 17012-901, SP, Brazil. [email protected].
- Department of Periodontics and Oral Medicine, School of Dentistry, University Michigan, Ann Arbor, MI, USA.
- Department of Orthodontics, School of Dentistry, Federal University of Rio de Janeiro, Rio de Janeiro, RJ, Brazil.
- Division of Orthodontics, School of Dentistry, University of Minnesota, Minneapolis, MN, USA.
- Division of Orthodontics and Division of Oral and Maxillofacial Radiology, School of Dentistry, Universidad Científica del Sur, Lima, Perú.
- Division of Oral and Maxillofacial Radiology, School of Dentistry, Universidad Nacional de Colombia, Bogotá D.C, Bogotá, Colombia.
- Division of Orthodontics, School of Dentistry, Universidad Nacional de Colombia, Bogota D.C, Colombia.
- Department of Orthodontics, Arthur Dugoni School of Dentistry, University of the Pacific, San Francisco, CA, United States of America.
- Department of Orthodontics, Bauru Dental School and Hospital of Rehabilitation of Craniofacial Anomalies University of São Paulo, Bauru, SP, Brazil.
- Department of Orthodontics, School of Dentistry, Universidade Federal Fluminense, Niterói, Brazil.
- Adams School of Dentistry, University of North Caroline, Chapel Hill, NC, USA.
Abstract
This study aimed to validate an artificial intelligence (AI)-based automated image analysis for three-dimensional (3D) characterization of impacted canine position. In addition, it compared clinical treatment plans developed using conventional orthodontic records versus with the aid of the AI-generated diagnostic report. A sample of 228 cone-beam computed tomography (CBCT) scans for patients with impacted maxillary canines was retrospectively collected. AI models were used to automatically orient scans and identify anatomical landmarks to quantify impaction severity. The accuracy of the AI approach was validated by comparing its assessments with manual measurements performed by expert clinicians. Clinical applicability was assessed by comparing diagnoses made by clinicians using conventional records to those made using AI-based diagnostic system. A total of 316 impacted canines were assessed, with 20.9% classified as buccal, 31.3% as bicortical, and 47.8% as palatal. The AI-based diagnostic system accurately identified the position of impacted canines, achieving a mean detection success rate of 96.2%. Palatally impacted canines tend to have the crown closer to the midpalatal plane, reduced vertical distance to the occlusal plane, and greater pitch and roll angles compared to bicortically and buccally impacted canines (P<0.001). Overlap with adjacent teeth was the primary contributor to the AI-generated severity index (81.11%). The AI-based diagnostic system report significantly influenced clinicians' decisions, particularly patient education (72.31%) and biomechanics (51.15%). The AI-based diagnostic system accurately characterized impacted canine position and severity, demonstrating the effectiveness of AI in enhancing diagnosis for impacted canine. The automated diagnostic report provides objective measurements that enhance clinicians' understanding of impaction complexity, improve patient communication, and support more informed biomechanical and treatment-planning decisions.