人工神经网络在肾小球滤过率估算中的应用研究 - 第三军医大学学报.doc

人工神经网络在肾小球滤过率估算中的应用研究 李宁山1，刘迅1，2，吴效明1，黄岳山1，娄探奇2 （510006广州，华南理工大学生物医学工程系1；510630 广州，中山大学附属第三医院肾内科2） [摘要] 目的 建立一个适用于中国慢性肾脏病人群的肾小球滤过率估算模型基于人体体征及估算肾小球滤过率。采用人工神经网络方法中的，基于562例训练样本集建立模型，在独立的269例验证样本集中验证模型性能，与传统的统计学回归方法得到的GFR估算经验方程比较。与经验方程相比，神经网络模型具有更高的准确性（P0.05）。人工神经网络作为常用的机器学习方法之一，应用于生物医学信息处理时，比传统统计学方法具有更大的优势，利用该方法建立的肾小球滤过率估算模型具有更好的估算精度Applied research of artificial neural network on estimating glomerular filtration rate Li Ningshan1, Liu Xun1,2, Wu Xiaoming1, Huang Yueshan1,Lou Tanqi2 （Department of Biology Engineering, South China University of Technology, Guangzhou, 510006；2Division of Nephrology, Department of Internal Medicine, The Third Affiliated Hospital of Sun Yet-sun University, Guangzhou, 510630,China） [ABSTRACT] Objective: To build a model for estimating glomerular filtration rate of Chinese patients with chronic kidney disease based on serum physiological parameters and demographic characteristics. Methods: GRNN, one approach of artificial neural network, was applied to build the model based on 562 training data set, and the performance of the model was validated on 269 validation data set, then it was compared with empirical equations derived from traditional regression method of statistics. Results: Compared with empirical equations, the performance of artificial neural network was better in accuracy with statistically significant differences (P0.05). Conclusion: Experimental result indicated approach of artificial neural network, as a common machine learning method, was superior to traditional statistics method in biomedical information processing. The model for estimating glomerular filtration rate was more accurate so that it point out a way of further study for even better model. [Key words] chronic kidney disease; glomerular filtration rate; artificial neural network Supported by the General Program of National Natural Science Foundation of China,Chinas postdoctoral scientific foundation fourth batch of special funded project(201104335),