Chen Xin, Zhang Zheng-Guo, Feng Kui, Chen Li, Han Shao-Mei, Zhu Guang-Jin
Department of Biomedical Engineering, Institute of Basic Medical Sciences, Chinese Academy of Medical Sciences, Peking Union Medical College, Beijing 100005, China.
Sheng Li Xue Bao. 2011 Aug 25;63(4):377-86.
The aim of this study is to develop backpropagation neural networks (BPNN) for better prediction of ventilatory function in children and adolescents. Nine hundred and ninety-nine healthy children and adolescents (500 males and 499 females) aged 10-18 years, all of the Han Nationality, were selected from Inner Mongolia Autonomous Region, and their heights, weights, and ventilatory functions were measured respectively by means of physical examination and spirometric test. Using the approaches of BPNN and stepwise multiple regression, the prediction models and equations for forced vital capacity (FVC), forced expiratory volume in one second (FEV1), peak expiratory flow (PEF), forced expiratory flow at 25% of forced vital capacity (FEF25%), forced expiratory flow at 50% of forced vital capacity (FEF50%), maximal mid-expiratory flow (MMEF) and forced expiratory flow at 75% of forced vital capacity (FEF75%) were established. Through analyzing mean squared difference (MSD) and correlation coefficient (R) of the ventilatory function indexes, the present study compared the results of BPNN, linear regression equation based on this work (LR's equation), prediction equations based on the studies of Ip et al. (Ip's equation) and Zapletal et al. (Zapletal's equation). The results showed, regardless of sex, the BPNN prediction models appeared to have smaller MSD and higher R values, compared with those from the other prediction equations; and the LR's equation also had smaller MSD and higher R values compared with those from Ip's and Zapletal's equations. The coefficients of variance (CV) for FEF50%, MMEF and FEF75% were higher than those of the other ventilatory function parameters, and their increasing percentages of R values (ΔR, relative to R values by LR's equation) derived by BPNN were correspondingly higher than those of the other indexes. In sum, BPNN approach for ventilatory function prediction outperforms the traditional regression methods. When CV of a certain ventilatory function parameter is higher, the superiority of BPNN would be more significant compared with traditional regression methods.
本研究旨在开发反向传播神经网络（BPNN），以更好地预测儿童和青少年的通气功能。从内蒙古自治区选取了999名10 - 18岁的健康儿童和青少年（500名男性和499名女性），均为汉族，分别通过体格检查和肺活量测试测量了他们的身高、体重和通气功能。采用BPNN和逐步多元回归方法，建立了用力肺活量（FVC）、一秒用力呼气容积（FEV1）、呼气峰值流速（PEF）、用力肺活量25%时的用力呼气流量（FEF25%）、用力肺活量50%时的用力呼气流量（FEF50%）、最大呼气中期流速（MMEF）和用力肺活量75%时的用力呼气流量（FEF75%）的预测模型和方程。通过分析通气功能指标的均方差（MSD）和相关系数（R），本研究比较了BPNN、基于本研究的线性回归方程（LR方程）、基于Ip等人研究的预测方程（Ip方程）和Zapletal等人研究的预测方程（Zapletal方程）的结果。结果表明，无论性别，与其他预测方程相比，BPNN预测模型的MSD较小，R值较高；与Ip方程和Zapletal方程相比，LR方程的MSD也较小，R值较高。FEF50%、MMEF和FEF75%的变异系数（CV）高于其他通气功能参数，BPNN得出的这些指标的R值增加百分比（ΔR，相对于LR方程的R值）相应高于其他指标。总之，BPNN用于通气功能预测的方法优于传统回归方法。当某个通气功能参数的CV较高时，与传统回归方法相比，BPNN的优势将更加显著。
