Artificial Intelligence Versus Human Intelligence in Presurgical Implant Planning: A Preclinical Validation.

Abstract

Objectives: To validate an innovative artificial intelligence (AI)-driven tool for automated virtual implant placement by comparing its accuracy, implant dimension selection, time efficiency, and consistency with a human intelligence (HI)-based approach for single posterior tooth replacement.

Materials and methods: A dataset of 50 time-matched cone-beam computed tomography and intraoral scans with a single missing posterior mandibular tooth was selected to validate a pre-trained AI model for virtual implant placement against a HI-based approach. A quantitative comparison of implant location and implant dimension selection was conducted between AI and HI, and a qualitative three-dimensional evaluation was conducted by three implant dentistry specialists using a visual analog scale and a Turing test to assess and distinguish between AI and HI. Additionally, time consumption and consistency were evaluated.

Results: Experts found that approximately 89% of AI-planned and 93% of HI-planned implants were clinically acceptable, with the planning method unidentifiable in 58% of AI cases. AI selected implant dimensions of 11.7 mm (1.3) in length and 4.0 mm (0.3) in diameter, close to experts' selections of 11.5 mm (1.3) and 4.2 mm (0.4). AI was over twice as fast, reducing planning time to 187 s (34) compared to 406 s (68) for HI (p < 0.0001), and demonstrated high consistency with a median surface deviation (MSD) of zero, while intra- and inter-operator MSDs were 0.33 mm (0.14) and 0.56 mm (0.34), respectively (p < 0.0001).

Conclusions: Artificial intelligence is reliable for virtual implant placement in missing mandibular (pre)molars, producing clinically acceptable plans comparable to human experts while operating faster and much more consistently than implant clinicians.