Application of Artificial Intelligence in Predicting Soft-Tissue Aesthetics and Patient Satisfaction after Maxillofacial Surgery

Abstract

Background: Predicting the postoperative facial appearance is an important issue to consider in the planning of maxillofacial surgery as the soft-tissue changes may impact aesthetics and patient satisfaction. Although artificial intelligence (AI) shows promise in predicting facial morphology after surgery, more research is needed to correlate objectively documented soft-tissue changes, with patients' satisfaction.

Objective: The purpose of this study was to assess the accuracy of this AI model in predicting aesthetic soft tissue changes after surgery and to establish the correlation between the prediction accuracy and patient satisfaction after maxillofacial surgery.

Methods: A prospective observational study was carried out on 220 patients between 20-40 years of age who were scheduled for maxillofacial surgery that involved a soft tissue or facial aesthetic component. Participants were recruited by consecutive sampling, based on predetermined criteria. A face-based AI prediction model was trained with pre-operative standardized facial photos and 3D facial scans and various other demographic and surgical parameters. Postoperative facial predictions using AI were compared with the standard postoperative three-dimensional facial scans after the specified healing period.After the specified healing period, AI-generated postoperative facial predictions were compared to standard postoperative three-dimensional facial scans. The mean absolute error (MAE), root mean square error (RMSE), landmark displacement error and intraclass correlation coefficients (ICC) were used to assess the accuracy of predictions for selected soft-tissue landmarks: lower-lip projection, upper-lip position, chin projection, facial midline, nasal projection, facial width and lower-face proportions. The FACE-Q domains of facial appraisal and a global satisfaction score were used to evaluate patient-reported aesthetic outcomes. Subsequently, associations between the accuracy of the AI predictions and patients' satisfaction were investigated.

Results: The AI-based model achieved significant correlation between predicted and postoperative soft-tissue morphology. The accuracy of the prediction was evaluated for each facial landmark and each regional aesthetic characteristic, with a lower prediction error resulting in greater similarity between the facial outcome predicted by the AI and the actual postoperative facial outcome. Major changes in the shape of the profile and contour were found to be very consistent and those of the soft tissue somewhat more variable. Patient reported outcomes showed better facial appearance, satisfaction with lower-face and jawline, satisfaction with lips and overall aesthetic satisfaction. The analysis also examined whether patients with smaller AI prediction errors had higher satisfaction scores, to determine the relationship between the objective accuracy of the prediction by the AI and the subjective aesthetic perception by the patient.

Conclusion: The results of the study indicated that the use of AI-assisted prediction could enable individualized assessment of the aesthetic outcomes of the soft tissues after maxillofacial surgery. Three-dimensional prediction-error measurements were integrated with patient-reported aesthetic outcomes to create a patient-centred evaluation of the performance of AI-based facial prediction. This could help to provide better individualised counselling and evaluation during the postoperative phase, by associating predicted facial alterations with the patient's aesthetic experience.