Development of a preoperative accuracy prediction model using machine learning for implant placement in static-guided surgery: retrospective observational study

Purpose This study aimed to develop a machine learning model capable of preoperatively predicting three-dimensional implant placement errors at the implant apex in static-guided surgery and to identify the clinical features associated with placement accuracy.
Methods Clinical data partially derived from a previous observational study were analyzed. In total, 181 patients and 480 implants placed using fully static-guided surgery were included in this study. The outcome variable was defined as three-dimensional implant placement error at the implant apex relative to the preoperative simulation, dichotomized as less than 0.5 mm or ≥ 0.5 mm. Twenty-one clinical and radiographic factors previously suggested to influence the placement accuracy were used as explanatory variables. The feature importance was evaluated using three gradient boosting decision tree models. Furthermore, a stacking model combining multiple classifiers was constructed, and the classification performance was assessed using ten-fold cross-validation.
Results The feature importance analysis identified 12 features associated with implant placement errors. The stacking model demonstrated superior classification performance compared to individual classifiers. The true positive rate was 0.73, false negative rate was 0.27, false positive rate was 0.14, and true negative rate was 0.86.
Conclusions The proposed stacking model correctly classified 86% of cases with implant placement error less than 0.5 mm and 73% of cases with implant placement error of ≥ 0.5 mm. These findings suggest that the proposed model may support the preoperative evaluation of implant placement accuracy in static-guided surgeries.