Predicting the patient-specific centre of rotation for total temporomandibular joint replacement using statistical shape modelling.

This study was performed to explore the use of statistical shape modelling (SSM) to predict the centre of rotation (CR) in patients undergoing total temporomandibular joint replacement (TMJ-TJR), in order to facilitate more accurate prosthesis design. In this retrospective study, a SSM of mandible halves was constructed using preoperative segmented three-dimensional models from cone-beam computed tomography scans, split into 388 mandible halves. The model was evaluated through compactness and generalization curves and by examining the principal components (PCs) of shape variations. A dataset of 40 mandible halves with four-dimensional optical tracking-derived CR measurements was fit to the SSM. Ridge regression was applied to predict patient-specific CR locations from SSM-derived PC scores. The predictions were evaluated using the root mean square error (RMSE) by comparing them to the true CR values. The SSM explained 71% of the total variation with the first five PCs. Increasing the number of PCs improved model accuracy. The regression model for predicting patient-specific CRs achieved a mean RMSE of 11.63 ± 11.01 mm, outperforming a naïve model as used in clinical practice that assumes a fixed CR location, which resulted in an RMSE of 31.88 ± 21.31 mm. A SSM of mandible halves to predict patient-specific CRs, capturing a significant portion of side-specific anatomical variability, was successfully developed. The regression analysis demonstrated a strong relationship between mandible shape, CR location, and corresponding PCs, laying a solid foundation for advancing the precision of TMJ-TJR procedures.