人工神经网络在预测输尿管结石自行排出中的应用研究  

Application of Artificial Neural Network in the Prediction of Spontaneous Ureteral Calculus Passage

作　　者：曾凯[1] 李应龙[1] 钱彪[1] 李强[1] 倪钊[1] 王新敏[1] 王文晓[1] 任之尚 王勤章[1] 

机构地区：[1]石河子大学医学院第一附属医院泌尿外科,新疆石河子市832008

出　　处：《中国全科医学》2016年第20期2430-2434,共5页Chinese General Practice

基　　金：国家科技支撑计划子课题(2013BAI05B05);新疆生产建设兵团医药卫生重点项目科技援疆专项(2014AB052)

摘　　要：目的运用人工神经网络建立输尿管结石自行排出的预测模型,并转化成临床应用。方法选取2013年1—8月在石河子大学医学院第一附属医院泌尿外科门诊就诊的输尿管结石患者225例。经保守排石治疗4周后,复查泌尿系B超或CT判断结石是否排出,并将患者分为排石组和未排石组。通过单因素分析筛选出影响结石排出的因素,将这些因素作为预测参数建立人工神经网络,并对68例测试集样本进行预测。绘制预测拟概率的ROC曲线,并计算ROC曲线下面积评价预测效能。结果排石组141例,未排石组84例。单因素分析结果显示两组患者性别、体质指数、膀胱刺激征、侧别、肾盂积水、尿p H值、血尿、淋巴细胞计数比较,差异均无统计学意义(P>0.05);两组患者年龄、疼痛程度、结石直径、结石位置、血白细胞计数、中性粒细胞计数、中性粒细胞分数、淋巴细胞分数、C反应蛋白水平比较,差异均有统计学意义(P<0.05)。运行人工神经网络,输入层共建立9个神经元。系统自动体系构建两个隐含层,输出层有1个神经元。预测变量重要性前3位分别是:结石直径(0.20)、C反应蛋白(0.18)、年龄(0.12)。运用建立成功的人工神经网络对68例测试集样本进行预测,结果显示,测试集样本的灵敏度、特异度和总准确率分别为93.3%、60.9%和82.4%,ROC曲线下面积为0.868〔95%CI(0.774,0.962)〕。结论人工神经网络预测输尿管结石能否排出有较高的准确率,可辅助临床医师为患者制定安全、合理的治疗方案。Objective To establish the prediction model of the spontaneous ureteral calculus passage by applying artificial neural network and put it into clinical application. Methods From January to August in 2013, we enrolled 225 patients with ureteral calculus who received treatment in the Department of Urology, the First Affiliated Hospital of Medical College,Shihezi University. After 4- week medical expulsive treatment,the status of calculus was examined by urinary tract ultrasound or CT, and patients were divided into two groups： calculus removed group and non calculus removed group. By univariate analysis, influencing factors for removing calculus were selected and were applied as predictive parameters in the establishment of artificial neural network prediction model, and the model was used to make prediction on 68 testing samples. ROC curve was made to predict quasi- probability,and the AUC was calculated to predict efficacy. Results There were 141 patients in calculus removed group and 84 patients in non calculus removed group. Univariate analysis showed that the two groups were not significantly different in gender,BMI,bladder irritation symptoms,lesion side,hydronephrosis,urine p H value,hematuresis and lymphocyte count（ P 〉0. 05）; the two groups were significantly different in age,pain degree,calculus size,position of calculus,leucocyte count,neutrophil count,neutrophil percentage,lymphocyte percentage and C- reactive protein level（ P〈 0. 05）. Artificial neural network was operated with all together 9 neurons in the input layer. Two hidden layers were established in the automatic system,and there was one neuron in the output layer. The first three predictive variables in importance were calculus size（ 0. 20）,C- reactive protein level（ 0. 18） and age（ 0. 12）. The neural network models that were successfully built were applied in the prediction of 68 testing samples,and the results showed that the sensitivity,specificity and total accuracy rate of artificial neural network model were 93.

关 键 词：输尿管结石 神经网络(计算机) 药物排石治疗 预测 

分 类 号：R693.4[医药卫生—泌尿科学] R319[医药卫生—外科学]