Data-efficient deep learning surrogate modeling for finite element prediction in dental implant biomechanics

Finite element analysis (FEA) is widely used to evaluate dental implant biomechanics but is computationally demanding. This study explores deep learning surrogate models for accurate stress prediction with limited data. A neural network was pre-trained on standard implant simulations (N = 3000) and adapted to narrow-diameter implants in atrophic bone (N = 200). Negative transfer occurred with extremely limited data (N = 20) due to biomechanical domain shift. However, both transfer learning and standard training achieved high accuracy at N ≥ 60 (R² = 0.94). This framework enables rapid, data-efficient biomechanical prediction for dental implant modeling.