Utilization of Machine Learning for the Objective Assessment of Rhinoplasty Outcomes

Machine Learning started to provide solutions to various challenges in many fields, including medicine. The objective assessment of rhinoplasty results has been a challenge since the assessment of beauty is subjective in nature. This study explores if Machine Learning can be used to accomplish the complex task of objective evaluating the outcome evaluation and automated scoring for rhinoplasty. We introduce a methodology to map the aesthetics of visual appearance to the quantified measurements of pre-surgery, planned outcome, and post-surgery using machine learning. To develop the methodology, we generated synthetic 3D models utilizing artificial intelligence tools and applied various nasal deformities to simulate the pre-surgery, planned outcome, and post-surgery scans of rhinoplasty patients. The simulated outcomes were scored by reviewing the 3D visuals and corresponding measurements to prepare the training data for machine learning models. AutoGluon AutoML framework is used to generate the best-performing machine learning model. We successfully developed machine learning models with accuracy between 82% to 88% depending on the scoring method. We also identified the measurements that are highly influential in determining the scores. This is the first study that correlates the visual appearance and quantitative facial measurements of simulated rhinoplasty outcomes. The results suggest that an AI-based objective rhinoplasty outcome scoring tool is possible when machine learning algorithms are trained using consensus scores along with patients' pre-surgery, planned, and post-surgery measurements. This study introduces a methodology regarding how to map the aesthetics of visual appearance to the quantified measurements of pre-surgery, planned outcome, and post-surgery using machine learning.