The Simulation Of Facial Plastic Surgery Based On Machine Learning

Facial plastic surgery has become very popular in recent years. With peo-ple's increasing pursuit of beauty and improvement of surgical techniques, plastic surgery has been accepted by more and more people. Also, a large number of cases have been accumulated in hospital. Usually, patients will require knowing the post-operative appearances beforehand, but it is incon-venient and inefficient to make physician-patient communication about issues of the surgery. Accumulated cases are only showed to patients in order to en-hance their knowledge of the surgery, which cannot make a definite and in-tuitive impression on them. Therefore, there is a strong need for a method to provide intuitive results after the operation by both surgeons and patients.With the increasing combination of computer technology and medicine, computer-aided simulation of plastic surgery has become a very important research topic. In recent years, there have been many related studies and ex-periments in this field at home and abroad, most of which pay attention to the simulation of bone operation. However, this kind of operation is very com-plicated and inconvenient.In this paper, a novel method for plastic surgery prediction based on ma-chine learning is presented. First of all, features that serve as input for train-ing and prediction are extracted from pre- and post-operative facial photos of patients; then one method using Support Vector Regression for the prediction of some main distances of local region and the other method using K Nearest Neighbor for the prediction of global difference between pre- and post-operative results are presented; in order to output an intuitive result, 2D and 3D visualization are performed based on quantized prediction data.Details of each step are described in experiment part and criteria are in-troduced to evaluate results of each step. Two standards, Maximum Closest Distance and Mean Closest Distance, are employed to compare prediction results. Cases of mandible reduction and cheekbone reduction are chosen in experiments, whose results show that the method using K Nearest Neighbor performs better than the one using Support Vector Regression in light of the both criteria. The average difference between prediction results and real post-operative results is about 5 pixels for facial region of about 420*320. 2D prediction results can serve as a good auxiliary tool for both doctors and pa-tients. As for three dimensional application, the prediction results cannot fit the real post-operative results accurately yet. The average difference between prediction result and the real result is 4.91mm. Therefore, it still cannot be applied in real clinical surgical planning by surgeons.