Accuracy of a Commercially Available AI Software for Immediate Implant

Although immediate implant placement is accepted as time efficient treatment method, the need for three-dimensional evaluation of alveolar socket morphology, cortical bone thickness and available bone volume is still crucial for primary stability and long term osseointegration. Evaluation and measurement of the conventional radiographic data and manual CBCT measurements are time consuming and experience-dependent. This study assessed the accuracy of a commercially available artificial intelligence software for immediate implant placement planning by comparing the AI-made measurements with conventional manual measurements by experienced implantologists. CBCT images of 318 patients (412 extraction sites) were evaluated in a retrospective study from 2019 to 2024. Standardized protocols (0.2 mm voxel size, 90 kVp, 8 mA) were used for all scans. The AI software automatically divided alveolar sockets, measured the thickness of buccal bone, socket dimensions and suggested implant size and angulation. Manual CBCT measurements were performed independently by two implantologists and the ground truth measurements were the intraoperative measurements. The Mean Absolute Error (MAE) and Intraclass correlation coefficient (ICC) were measured to assess the accuracy, and Bland–Altman analysis was used. The AI software resulted in MAE of 0.31 mm, 0.28 mm, and 0.34 mm in the buccal bone thickness, mesiodistal socket width and apicocoronal depth, respectively. There was excellent agreement between the measurements taken using the AI (ICC 0.89-0.93) and during surgery (ICC 0.93-0.96). The accuracy of the AI-based implant recommendations was 81.4%, while the surgeons were 87.2% accurate when surgically placing the implants. There was no systematic bias: Bland–Altman showed a mean difference of 18.7 – 4.2 = 14.5 minutes (95% limits of agreement: ±0.6 mm). The results show that AI-supported implant planning achieves clinically acceptable precision and significantly speeds up the analysis time, making it applicable to high-volume clinical practices.