摘要
目的:通过分析厂家提供的计算小切口角膜基质透镜取出(SMILE)手术角膜切削厚度值参考的标准数据及患者历史临床数据,构建多层感知器(MLP)神经网络模型,用于全飞秒SMILE手术角膜切削厚度的精准预测。方法:对医院SMILE手术共计1127例临床患者数据进行仿真验证,构建MLP神经网络模型,由球镜度数、柱镜度数、角膜曲率和微透镜直径4个影响因素组成输入向量,角膜切削厚度作为输出向量,将神经网络模型进行训练并保存,用于角膜切削厚度的预测。结果:仿真试验表明,多元线性回归方法计算的平均绝对误差(MAE)为5.791,均方误差(MSE)为60.966;MLP神经网络方法计算的平均绝对误差(MAE)为0.491,均方误差(MSE)为0.554,因此使用MLP神经网络效果更优。结论:构建的MLP神经网络模型实现了角膜切削厚度与其影响因素之间的非线性关系描述,MLP神经网络训练完成后可用于眼科诊疗过程中角膜切削厚度的快速计算,实现全飞秒SMILE手术预诊断功能,提高诊疗效率。
Objective:To construct a multi-layer perceptron(MLP)neural network so as to use in the accurate prediction of corneal cutting thickness in small-incision lenticule extraction(SMILE)surgery with entire femtosecond by analyzing the referenced standard data that were used to calculate corneal cutting thickness of SMILE surgery,which were provided by manufacturer,and the historical data of patients.Methods:The data of 1127 patients who underwent SMILE surgery were used to conduct verification for simulation effect.In the constructed MLP neural network model,4 influence factors included spherical degree,cylindrical degree,corneal curvature and the diameter of micro lens were composed of input vector,and the corneal cutting thickness was used as output vector.The neural network model was trained and saved to predict the corneal cutting thickness.Results:The results of simulation experiments showed that the mean absolute error(MAE)calculated by the multiple linear regression method was 5.791,and the mean square error(MSE)was 60.966.The MAE and MSE were 0.491 and 0.554 as the calculation of MLP neural network method,respectively.Therefore,the effect of using MLP neural network was better.Conclusion:The constructed MLP neural network realized the description of nonlinear relationship between the corneal cutting thickness and the influence factors of that,which can be used in the rapid calculation of corneal cutting thickness after the train of MLP neural network was completed,so as to realize the function of preemptive diagnosis of SMILE surgery with entire femtosecond and increase the efficiency of diagnosis and treatment.
作者
汤福南
张可
竺明月
杨春花
张晖
汪缨
袁冬青
TANG Fu-nan;ZHANG Ke;ZHU Ming-yue(不详;Department of Clinical Medical Engineering,The First Affiliated Hospital with Nanjing Medical University(Jiangsu Province Hospital),Nanjing 210029,China)
出处
《中国医学装备》
2022年第9期1-5,共5页
China Medical Equipment
