Study On Machine Learning Assisted SMILE Surgery Design

ObjectiveIn this study,for the over-correction or under-correction problems after SMILE surgery,the machine learning technology is used to combine the existing SMILE corneal refractive surgery measurement data and surgical parameters,analyze the important factors that affect the surgical results,and build a prediction model to improve surgical accuracy and predictability.At the same time,clinical controlled experimental studies were conducted to confirm the effectiveness of the model.MethodsThe whole patients are divided into two parts.In the first part,based on 865successful surgical cases in the SMILE surgical patient database,the patient’s personal basic information data,human eye refractive data,corneal morphology data and surgical parameter data were retrospectively screened.The tree algorithm analyzes and digs out the important factors that affect the operation result and attempts to use the neural network algorithm to construct the prediction model.In the second part,a prospective controlled study from the refractive surgery center of Tianjin Eye Hospital with a total of 3000 patients undergoing SMILE surgery was divided into two groups,surgeon group(group A)and machine learning group(group B),applying the Nomogram value prediction algorithm model computerized in the first part of the computer learning group of surgical parameters,and comparing the surgical results of the two groups for 6 months postoperative follow-up to evaluate the accuracy of the algorithm model Sex,effectiveness,stability and predictability.Results1.Retrospectively analyzed the clinical diagnosis parameters of 865 eyes of successful operation cases,and found the weight of important factors that affect the operation results after information gain calculation.Comparing different algorithms,it is found that the model based on multi-layer perception neural network is better.After 10fold cross-validation,the average prediction accuracy of the model is 95.14%.At the same time,the information gain algorithm used in this study not only found that some of the clinically confirmed parameters have an impact on the surgical results,but also found that the new parameters will also affect the SMILE surgical results.2.Using an algorithm model constructed by multi-layer perception neural network,clinical control found that it has more consistent accuracy with the surgeon.In addition,the safety indexes of the patients in groups A and B were 0.989±0.056 and0.997±0.259(P>0.05),indicating that the machine learning group is same with the surgeon group.The effectiveness indexes were 1.162±0.278 and 1.174±0.300(P<0.05).There were statistical differences between the two groups.The effectiveness index of the machine learning group was better than that of the surgeon group.For predictability,the correlation between the predicted corrected equivalent spherical power and the actual corrected equivalent spherical power of the two groups of patients is R~2=0.9134 and R~2=0.9162(P>0.05),respectively.The display is more consistent.ConclusionBased on machine learning technology,this study analyzes and calculates the influencing factors of SMILE surgery from the big data of refractive surgery,and shows better performance than clinical experts in the task of designing the SMILE surgery Nomogram value parameter of myopia patients.It can realize the auxiliary task of clinical diagnosis and treatment of SMILE surgery,and further improve the accuracy of surgery.