人工神经网络在糖尿病并发症预测中的应用

Application of artificial neural network in diabetic complication prediction

目的通过人工神经网络的科学计算方法来预测糖尿病患者并发症的发展趋势。方法分析100例糖尿病患者的临床资料及其空腹血糖（Glu）、餐后2h血糖（P2hBG）、糖化血红蛋白（HbA1c）、低密度脂蛋白胆固醇（LDL—C）4个指标，然后根据LDL—C正常、异常以及有无并发症将糖尿病患者分为两组（各50例），并将每一组归纳为一个集合，应用MATLAB6．5软件建立糖尿病并发症预测模型，同时随机抽取18例临床确诊的糖尿病患者，将其Glu、P2hBG、HbA1c、LDL-C数据作为输入值进行模型验证，以检测网络的拟合效果。结果人工神经网络预测与随访情况比较，预测糖尿病并发症的灵敏度为66．7％，特异度为100．0％，准确度为83．3％。结论利用人工神经网络进行糖尿病并发症预测，能取得良好的预测效果。

Objective To identify diabetic complication tendency through artificial neural network science calculation. Methods The clinical data of 100 patients with diabetes mellitus, fast blood glucose （Glu）, 2 hours postprandial blood glucose （P2hBG）, HbA1c and LDL-C were analysed. According to LDL-C level and diabetic complication, the patients were divided into two groups. MATLAB6.5 software was employed to eatablish the prediction models of diabetic complication. Data of Glu, P2hBG, HbAlc and LDL-C from 18 cases of diabetic mellitus were selected at random and employed as input values to confirm artificial neural network fit effect. Results The sensitivity, specificity and accuracy of diabetic complication prediction were 66. 7%, 100.0% and 83. 3% respectively. Conclusion Artificial neural network in diabetic complication prediction can gain good effect.