神经网络模型和多元线性回归预测肾结石CT值的比较

Comparison of neural network model and multiple linear regression in predicting CT value of kidney stones

目的:比较神经网络模型(NNM)和多元线性回归(MLR)预测肾结石CT值的能力。方法:回顾性搜集2018年5月至2018年12月我院诊断为肾结石的患者121例(男95例,女26例),采用随机表法按3:1分为训练样本和验证样本。所有患者均进行CT平扫成像,比较两种模型的精确程度。结果:训练样本中,NNM计算得到C’与C的相关系数高于MLR;验证样本中,NNM得到C’与C的相关系数也高于MLR。训练样本中,NNM得到的MAE、MRE和RMSE均低于MLR,差异有统计学意义;验证样本中,NNM得到的MAE(17.351)、MRE(3.822)和RMSE(3.116)也低于MLR,两者间差异均无统计学意义。结论:NNM较MLR能更好地预测肾结石的CT值。

Objective To compare the ability of neural network model(NNM)and multiple linear regression(MLR)to predict CT value of kidney stones.Methods A total of 121 patients(95 males and 26 females)diagnosed with renal stones in our hospital from May 2018 to December 2018 were retrospectively collected and randomly divided into training samples and verification samples according to 3:1.All patients underwent plain CT imaging to compare the accuracy of the two models.Results In the training samples,the correlation coefficient between C'and C calculated by NNM was higher than that of MLR.In the verified samples,the correlation coefficient between C'and C obtained by NNM was also higher than that of MLR.In the training samples,MAE,MRE and RMSE obtained by NNM were all lower than MLR,with statistically significant differences.In the validation samples,MAE(17.351),MRE(3.822)and RMSE(3.116)obtained by NNM were also lower than MLR,with no statistically significant differences.Conclusion NNM can better predict CT value of kidney stones than MLR.