Improving the video imaging prediction of postsurgical facial profiles with an artificial neural network

Background/purpose

With advancements in computer technology, postsurgical video image simulations are becoming more frequently used in orthognathic surgery. Simulations can greatly affect decision making by patients and also provide information to surgeons and orthodontists. However, most of the current commercial video image prediction software is only suitable for patient education but is not precise enough for clinical communication and treatment planning. The purpose of this study was to evaluate and improve post-orthognathic surgery image predictions.

Materials and methods

In this retrospective study, 30 bimaxillary protrusion patients who underwent two jaw surgeries were recruited. Simulations were compared with the actual postsurgical facial profile. An artificial neural network (ANN) was used to improve the predictions.

Results

The lower lip was the least accurate point, and the prediction error on the sagittal plane was +4.0 mm. After applying the ANN to the input data, the prediction error was reduced to +0.3 mm with a > 80% improvement rate. The overall probability of the prediction errors being < 2 mm was 52% before improvement and 84.5% after improvement. Improvement rates of the average prediction errors on the sagittal and vertical planes were 43.9% and −6.6%, respectively.

Conclusion

With the help of an ANN, the accuracy and reliability of the postsurgical profile video image predictions were greatly improved to a clinically applicable and treatment planning level.

Key Words:  artificial neural network , improvement , orthognathic surgery , prediction , video image

No full text is available. To read the body of this article, please view the PDF online.