Development And Application Of An Artificial Intelligence System For Rhinoplasty Information Complementary Based On Deep Learning

Objective: Generative Adversarial Networks(GANs)is a key progress in Machine Learning in recent years,which greatly expands the application scope of Artificial Neural Networks(ANNs)and enables artificial intelligence to be creative.In plastic surgery,rhinoplasty is considered to have the highest dissatisfaction rate among all facial cosmetic surgeries due to its difficulty and abstract nasal morphology.The key factors to improve patients’ postoperative satisfaction are preoperative communication and preoperative simulation of nasal morphology between doctors and patients.In this study,we hope to combine GANs’ creativity to apply this algorithm to simulate nasal morphogenesis before rhinoplasty.Methods:1.More than 300 cases of rhinoplasty performed by the same doctor in the Department of Plastic and Aesthetic(Burn)Surgery of The Second Xiangya Hospital were collected.Preoperative and postoperative standard orthopedic photographs of frontal,right 45° and right 90° were used as the training data set.At the same time,because the purpose of the deep learning training set is to predict and simulate the nasal shape of rhinoplasty patients,which is in line with modern aesthetics,all the selected patients were screened and scored accordingly.2.After the preliminary test,the preoperative and postoperative photos of all patients were preprocessing to remove redundant information and redundant noise.The preoperative and postoperative photos of all patients were used as input and output respectively to conduct iterative training in the established network framework.3.In order to verify the rationality and effectiveness of the image generation algorithm in this paper,subjective evaluation method and objective evaluation method are adopted respectively.The objective evaluation method adopts peak signal-to-noise ratio measure(PSNR)and structural similarity(SSIM).4.To further verify the clinical usability of the preoperative rhinoplasty simulation of our algorithm,it was compared with the control group of the preoperative 3D simulation performed by Da Vinci Plastic Surgery,which is three-dimensional preoperative simulator currently used in the clinic.Results:1.A total of 323 female patients aged between 18 and 50 who underwent rhinoplasty by the same doctor in the Department of Plastic and Aesthetic(Burn)Surgery,Second Xiangya Hospital of Central South University were collected.Among them,234 had achieved 8 points,19 had achieved 9 points,and two had achieve 10 points from more than five plastic surgeons in the VAS.A total of 255 patients were included in this deep learning network training set.2.This paper designs an improved algorithm based on pix2 pix for the problems of poor generating effect and large difference between generated image and target image in supervised image generation algorithm.The innovation of the algorithm lies in the combination of overall matching discrimination networks and overall true and false discrimination networks to complete the image generation,which not only improves the defects of other supervised image generation algorithms,but also improves the accuracy of image generation.Finally,the nasal morphology was simulated successfully in some cases,and the generated morphology image was highly similar to the original postoperative morphology of the patients.3.The final image evaluation results show that the PSNR and SSIM values of the image algorithm in this paper are of high level.However,due to the lack of other nasal morphology generation algorithms,the results cannot be compared with those of other algorithms.For the supervisor’s evaluation,the shape of the generated external nose was basically consistent with the actual shape of the patient after rhinoplasty.Direct observation of the images with naked eyes could obviously show that the shape of the patient’s external nose was improved,and the shape of the nose was basically complete and clear,including the skin color and texture,which were natural and reliable.4.In comparison with the preoperative 3D simulation completed by INOVA 3D-EX Da Vinci Plastic Surgery system,it can be seen that the preoperative simulation of rhinoplasty generated by the algorithm in this study has a high similarity with the results of Da Vinci plastic surgery system,which proves that the preoperative simulation of the algorithm has great clinical application potential.It can be used in clinical practice after ethical review.Conclusion: This study realized the application of deep learning to complement and generate biological information in the field of Rhinoplasty,thereby achieving the goal of better surgical results.The realization of the algorithm was explored by taking the simulation of rhinoplasty as an example,and the results were compared with the three-dimensional preoperative simulation equipment in clinical application.With the development of biomaterials and tissue engineering,this research direction may have greater potential.At the same time,this study can be extended to other plastic surgery projects,such as ear reconstruction,facial reconstruction,etc.And it is expected to gain greater value.