论文部分内容阅读
目的:建立预测ESWL治疗肾结石成功率的人工神经网络,并转化为临床应用。方法:2008年1月~2010年2月接受ESWL治疗的肾结石患者325例,将治疗前的临床资料共10项(性别、尿路刺激症、血尿、肾绞痛、结石位置、结石患侧、年龄、体重指数、病程时间、结石大小)纳入预测参数,建立人工神经网络。绘制预测拟概率的ROC曲线,计算曲线下面积(area under curve,AUC),计算约登指数,结合临床要求,得到适宜的概率分界值。结果:人工神经网络得到预测参数重要性的前5位依次是结石大小、病程时间、血尿、结石位置、体重指数。ANN的AUC为0.856,95%可信区间为0.774~0.917,与AUC为0.5进行显著性检验,P<0.001。确定概率分界值为0.595时,ANN的敏感度和特异度达到较为理想状态,分别为92%和60%。结论:人工神经网络预测ESWL单次治疗肾结石成功率有较好的准确性,选择恰当的概率分界值,可提高人工神经网络的预测效能。
OBJECTIVE: To establish an artificial neural network to predict the success rate of ESWL in the treatment of kidney stones and to translate it into clinical application. Methods: From January 2008 to February 2010 325 patients with ESWL were treated by ESWL. The clinical data before treatment were 10 (gender, urinary tract irritation, hematuria, renal colic, location of stone, side of stone , Age, body mass index, duration of disease, stone size) into the prediction parameters, the establishment of artificial neural network. The ROC curves of the predictive probability were plotted, the area under curve (AUC) was calculated, and the Youden index was calculated. According to clinical requirements, the appropriate probability cutoff values were obtained. Results: The top five parameters that are the most important parameters of artificial neural network are stone size, duration of disease, hematuria, location of stones, body mass index. The AUC of ANN was 0.856, with a 95% confidence interval of 0.774-0.917, a significant test with AUC of 0.5 (P <0.001). When the probability cutoff value was 0.595, the sensitivity and specificity of ANN reached the ideal state, which were 92% and 60% respectively. Conclusion: Artificial neural network is more accurate in predicting the success rate of ESWL single treatment of kidney stones. Choosing the appropriate probability cutoff value can improve the prediction efficiency of artificial neural network.