Agreement Between Artificial Intelligence-Based and Expert Consensus Implant Planning Using Cone-Beam Computed Tomography: A Retrospective Observational Study.
Authors
Affiliations (7)
Affiliations (7)
- Department of Dentistry, PKG Medical College and Hospital, Kolkata, IND.
- Department of Periodontics, Drs. Sudha and Nageswara Rao Siddhartha Institute of Dental Sciences, Gannavaram, IND.
- Department of Orthodontics and Dentofacial Orthopedics, Sri Hasanamba Dental College and Hospital, Hassan, IND.
- Department of Periodontics, Bharati Vidyapeeth (Deemed to be University) Dental College and Hospital, Navi Mumbai, IND.
- Department of Prosthodontics, K. M. Shah Dental College and Hospital, Sumandeep Vidyapeeth (Deemed to be University), Vadodara, IND.
- Department of Periodontics, Maharishi Markendeshwar College of Dental Sciences and Research, Maharshi Markandeshwar (Deemed to be University), Mullana, IND.
- Department of Computer Science, Jazan University, Jazan, SAU.
Abstract
Artificial intelligence (AI) is increasingly being incorporated into digital implant dentistry, but the quantitative validation of AI-generated implant plans against expert consensus remains limited. Therefore, this study aimed to evaluate the agreement between an AI-based implant planning system and an expert consensus plan using cone-beam computed tomography (CBCT) images of posterior edentulous sites. This cross-sectional, observational study included 50 posterior edentulous sites from 50 adult patients. Two experienced implantologists independently planned implant length, diameter, mesiodistal position, and angulation, and a consensus plan was established as the reference standard. The same datasets were analyzed using an AI-based implant planning system. Agreement was assessed using the intraclass correlation coefficient (ICC), mean absolute error (MAE), Bland-Altman analysis, Cohen's kappa, and predefined clinical tolerance limits. Categorical agreement was 86% for implant length, 90% for diameter, 84% for position, and 88% for angulation. The ICC values were 0.91 (95% CI, 0.85-0.95) for length, 0.87 (0.78-0.92) for diameter, 0.85 (0.75-0.91) for position, and 0.80 (0.68-0.88) for angulation. The mean differences between AI and consensus were 0.1 mm for length, 0.1 mm for diameter, -0.1 mm for position, and 0.7<sup>o</sup> for angulation; none were statistically significant (all p > 0.05). Clinically acceptable agreement was observed in 94%, 90%, 88%, and 82% of cases, respectively. AI-based implant planning demonstrated good-to-excellent agreement with expert consensus planning across implant dimensions and three-dimensional positioning parameters. The findings indicate substantial concordance between AI-generated and expert-derived planning recommendations under the conditions evaluated. However, agreement with expert planning should not be interpreted as evidence of clinical or surgical accuracy. Prospective validation against actual postoperative implant positions is required to determine clinical accuracy.