基于尿常规的机器学习辅助泌尿系统肿瘤筛查  

Assisted Screening of Urological Carcinoma by Deep Learning Based on Human Routine Urianlysis

作　　者：王正[1] 王占宇 王金申[2] 刘志[3] 季凯 金讯波[1] WANG Zheng;WANG Zhan-yu;WANG Jin-shen;LIU Zhi;JI Kai;Jin Xun-bo(Minimally Invasive Urology Cencer,Shandong Provincial Hospital affiliated to Shandong University,Jinan,Shandong,250014,China;Department of Gastrointestinal surgery,Shandong Provincial Hospital affiliated to Shandong University,Jinan,Shandong,250014,China;Department of Clinical Laboratory,Shandong Provincial Hospital affiliated to Shandong University,Jinan,Shandong,250014,China;Shandong helical matrix data technology Co.,Ltd,Jinan,Shandong,250014,China)

机构地区：[1]山东大学附属省立医院泌尿微创中心 [2]山东大学附属省立医院胃肠外科 [3]山东大学附属省立医院检验科 [4]山东螺旋矩阵数据技术有限公司

出　　处：《泌尿外科杂志（电子版）》2018年第2期13-16,共4页Journal of Urology for Clinicians(Electronic Version)

基　　金：山东省重点研发计划(No.2017G006007)基金支持

摘　　要：目的探索利用机器学习基于尿常规检查结果的数据,进行辅助筛查泌尿系统肿瘤。方法利用PTSVM算法对500位正常人和408位泌尿系统疾病患者的尿常规数据进行分析,找到其与泌尿系统恶性肿瘤的相关性。结果对于泌尿系统恶性肿瘤,通过5次交叉验证,机器学习的最优平均分类准确率达到了85. 78%。结论 PTSVM算法可以通过尿常规检查数据区分正常人和泌尿系统恶性肿瘤患者,表明该方法有望成为一种泌尿系统肿瘤辅助筛查手段。Objective To explore an assist method for the diagnosis of urological carcinoma based on routine urianlysis by deep learning. Methods The progressive transductive support vector machine （PTSVM） was applied to distinguish the carcinoma of urinary system by analyzing the data of routine urianlysis of 500 patients of normal persons and 408 patients of urological malignancies. Results The average accuracy rate of deep learning method through 5 cross validation is 85.78% for urological malignancies, there was no differences between SVM and ANN. Conclusions There is probability to classify normal and urinacy system malignancies patients by using deep learning method on routine urianlysis.

关 键 词：尿常规 机器学习 支持向量机 泌尿系统肿瘤 疾病筛查 

分 类 号：R737[医药卫生—肿瘤]