应用人工神经网络评估大剂量甲氨蝶呤化疗后的骨髓抑制

目的：研究建立人工神经网络模型用于评估大剂量甲氨蝶呤(HDMTX)化疗后的骨髓抑制程度,促进个体化用药。方法：收集180例急性淋巴细胞白血病(ALL)患儿行HDMTX化疗的临床资料。将所有资料随机分成两组,训练组(n=150)：以化疗后中性粒细胞总数(NEU)减少率为输出目标,采用遗传算法配合动量法训练后建立人工神经网络；测试组(n=30)：用建立的人工神经网络预测测试组患儿的NEU减少率,通过计算平均预测误差(MPE)、权重残差(WRES)、平均绝对预测误差(MAE)、平均预测误差平方(MSE)和均方根预测误差(RMSE)来验证模型。结果：人工神经网络的MPE 为(-2.05±7.41)%,WRES为(23.20±29.74)%,MAE为(6.12±4.53)%,MSE为(57.26±64.46)(%)₂,RMSE为 7.57%,有76.67%的病例相对预测误差在±20%以内。人工神经网络预测的准确度及精密度均优于多元线性回归模型(逐步回归法)。结论：本研究建立的人工神经网络预测性能较好,可用于预测HDMTX化疗后骨髓抑制程度以指导个体化用药。

Application of artificial neural network in evaluation of the bone marrow depression following high-dose methotrexate chemotherapy

Objective: To establish an artificial neural network (ANN) to evaluate the bone marrow depression following high-dose methotrexate (HDMTX) chemotherapy, and to facilitate individualized therapeutic regimens. Methods: Data obtained from 180 cases of children with acute lymphoblastic leukemia (ALL) during HDMTX treatment were divided into two groups randomly, as training group(n=150/data sets) and testing group(n=30/data sets). The decrease percent in NEU count post-HDMTX infusion was selected as the ANN output, which was the prediction marker of bone marrow depression. ANN was established after the network parameters were trained by using momentum method combined with genetic algorithm based on the training group data. The decrease percent in NEU count of testing group patients were predicted by ANN established, and the mean predicted error(MPE), weighted residuals (WRES), mean absolute prediction error(MAE), mean squared prediction error(MSE), root mean squared prediction error (RMSE) were calculated to assess the ANN model. Results: The assessed results of ANN were MPE (-2.05±7.41)%, WRES (23.20±29.74)%, MAE (6.12±4.53)%,MSE (57.26±64.46)(%)₂, RMSE 7.57%,respectively. There were 76.67% of relative prediction error within ±20%. The accuracy and precision of ANN were superior to those of multiple linear regression with stepwise method. Conclusions: The performance of ANN established in this study was good enough to predict the degree of bone marrow depression following HDMTX chemotherapy and optimize individualized saving regimens.