Does artificial intelligence predict orthognathic surgical outcomes better than conventional linear regression...

Area of Science:

• Orthodontics and Dentofacial Surgery
• Medical Artificial Intelligence
• Surgical Outcome Prediction

Background:

• Orthognathic surgery requires accurate outcome prediction for optimal results.
• Conventional prediction methods, such as multivariate multiple linear regression (MLR) and partial least squares (PLS), have limitations.
• The integration of artificial intelligence (AI) offers potential advancements in predictive modeling.

Purpose of the Study:

• To compare the predictive performance of an AI model (TabNet) against conventional methods (MLR, PLS) for orthognathic surgical outcomes.
• To identify specific soft-tissue landmarks where AI or conventional methods excel in prediction.
• To analyze factors influencing prediction accuracy in orthognathic surgery.

Main Methods:

• Analysis of lateral cephalograms from 705 patients undergoing combined surgical-orthodontic treatment.
• Utilized 254 input variables including skeletal and soft-tissue characteristics, and surgical repositioning.
• Compared prediction accuracy of MLR, PLS, and a TabNet-based AI model for 32 soft-tissue landmarks.

Main Results:

• Multivariate multiple linear regression (MLR) exhibited the lowest predictive performance.
• Partial least squares (PLS) outperformed AI in predicting 16 soft-tissue landmarks above the upper lip.
• AI demonstrated superior prediction for six landmarks in the lower mandible and neck region; 10 landmarks showed no significant difference.

Conclusions:

• AI prediction models do not universally surpass conventional methods in orthognathic surgery.
• Partial least squares (PLS) showed better prediction for upper lip landmarks, while AI excelled in lower mandibular and neck areas.
• A hybrid approach combining AI and conventional methods may enhance the prediction of orthognathic surgical outcomes.