基于预测输尿管结石自然排出的人工神经网络模型的建立及应用

发布时间：2018-03-29 08:45

本文选题：输尿管结石　切入点：神经网络　出处：《石河子大学》2017年硕士论文

【摘要】：目的输尿管结石是泌尿外科常见疾病,保守药物排石作为传统的非侵入治疗方案广泛被患者所接受,但是目前尚无一种可靠的预测输尿管结石自发排出的方法。为此,本研究拟运用人工神经网络技术建立输尿管结石自发排出的预测模型,并转化成临床应用。方法选取2013年1月至2013年8月间前来我院就诊的225例输尿管结石患者作为研究对象,所有患者须符合纳入及排除标准。收集患者的临床资料包括一般情况,实验室检查指标及影像学检查资料。通过保守排石治疗4周后复查泌尿系超声或CT判断结石是否排出,将所有患者分为结石排出组和未排出组。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预测模型,并应用该模型对测试集样本进行预测。绘制预测拟概率的ROC曲线,并计算曲线下面积评价预测效能。为进一步评价该模型的泛化能力,再次随机选取44例选择保守排石治疗的输尿管结石患者,运用计算神经网络模型预测结石自排结局,再次评价该模型的预测效能。结果排石组141例,未排石组84例。通过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pH值、血尿、淋巴细胞计数比较,差异均无统计学意义(P0.05);两组患者年龄、疼痛程度评分、血白细胞计数、中性粒细胞百分比、淋巴细胞百分比、中性粒细胞计数、C反应蛋白值、结石大小及位置在排石组与未排石组间比较差异有统计学意义(P0.05)。系统将225例样本按照7:3的的比例随机分成训练集157例(70%)和测试集68例(30%)分别用于模型的建立及测试。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。预测变量重要性位于前三位的是结石直径(0.20)、C反应蛋白值(0.18)及患者年龄(0.12),应用该模型对68名测试集样本进行预测,结果显示测试集样本的敏感度、特异度和总准确率分别为93.33%,60.87%和82.35%。ROC曲线下面积为0.868,[95%CI(0.774,0.962)]。对随机选取的44名患者输尿管结石自排结局进行了预测,根据随访结果,计算出人工神经网络模型预测的敏感度、特异度和总准确率为100.00%,64.29%和88.64%。结论人工神经网络模型能准确预测输尿管结石能否排出,可辅助临床医师为患者制定安全、合理的治疗方案。
[Abstract]:Objective Ureterolithiasis is a common disease in urology, conservative drug lithotomy is widely accepted by patients as a traditional non-invasive treatment, but there is no reliable method to predict the spontaneous discharge of ureteral calculi. In this study, artificial neural network (Ann) technique was used to establish a predictive model of spontaneous ureteral calculi excretion, which was transformed into clinical application. Methods 225 patients with ureteral calculi who came to our hospital from January 2013 to August 2013 were selected as the study objects. All patients should meet the criteria of inclusion and exclusion. The clinical data including general information, laboratory examination indexes and imaging examination data were collected. After 4 weeks of conservative lithotomy treatment, urinary tract ultrasound or CT was rechecked to determine whether or not the stones were excreted. All patients were divided into stone excretion group and non-excretion group. Factors affecting stone excretion were screened by single factor analysis, and these factors were used as predictive parameters to establish artificial neural network prediction model. The model is used to predict the sample of the test set, the ROC curve of the forecast quasi probability is drawn, and the area under the curve is calculated to evaluate the prediction efficiency. In order to further evaluate the generalization ability of the model, Forty-four patients with ureteral calculi who were treated with conservative lithotomy were randomly selected to predict the outcome of calculi self-drainage by using the computational neural network model and to evaluate the prediction effectiveness of the model again. The results of single factor analysis showed that there was no significant difference in sex, BMI, bladder irritation, side, hydronephrosis, urine pH value, hematuria and lymphocyte count between the two groups (P 0.05). Pain score, white blood cell count, neutrophil percentage, lymphocyte percentage, neutrophil count and C-reactive protein, The difference of stone size and location between the two groups was statistically significant (P 0.05). 225 samples were randomly divided into training set (157 cases) and test set (68 cases) to set up the model according to the proportion of 7:3. Try. Run the artificial neural network, Nine neurons were created in the input layer, and two hidden layers were constructed by the automatic system. There was one neuron in the output layer. The prediction variables in the first three places were stone diameter 0.20 C reactive protein (0.18) and patient age 0.12. The model was used to predict 68 test set samples, and the results showed the sensitivity of the test set samples. The specificity and the total accuracy were 93.330.87% and 0.868 under the 82.35%.ROC curve, respectively. The self-discharging outcome of 44 patients with ureteral calculi selected randomly was predicted. The sensitivity of artificial neural network model was calculated according to the follow-up results. Conclusion the artificial neural network model can accurately predict whether ureteral calculi can be excreted, and can assist clinicians to formulate safe and reasonable treatment schemes for patients with ureteral calculi. The specificity and total accuracy are 100.00,64.29% and 88.64.Conclusion the artificial neural network model can accurately predict whether ureteral stones can be discharged.
【学位授予单位】：石河子大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R693.4

【参考文献】