基于人体血液学检测的机器学习辅助泌尿系肿瘤筛查  

Assisted Screening of Urological Carcinoma by Deep Learning Based on Human Hematological Examination

作　　者：王正[1] 王金申[2] 刘志[3] 季凯 刘义庆[3] 金讯波[1] WANG Zheng;WANG Jin - shen;LIU Zhi;JI Kai;LIU Yi - qing;JIN Xun - bo(Minimally Invasive Urology Center,Provincial Hospital Affiliated to Shandong University,Jinan,Shandong,250014,China;Department of GastrointestinalSurgery,Shandong Provincial Hospital Affiliated to Shandong University,Jinan,Shandong,250014,China;Department of clinical laboratory,ShandongProvincial Hospital Affiliated to Shandong University,Jinan,Shandong,250014,China;Shandong Helix Matrix 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,山东济南250014 [2]山东大学附属省立医院胃肠外科,山东济南250014 [3]山东大学附属省立医院检验科,山东济南250014 [4]山东螺旋矩阵数据技术有限公司,山东济南250014

出　　处：《泌尿外科杂志（电子版）》2017年第4期9-14,共6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：目的探索应用人工智能机器学习算法单纯基于肝、肾功等血液学检查来辅助筛查泌尿系统肿瘤。方法分别利用支持向量机和神经网络算法对3136例正常人员和泌尿系统恶性肿瘤患者肝肾功数据进行分析,找到肝肾功数据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次交叉验证的最优平均分类准确率达到了92.05%,支持向量机与神经网络算法结果基本一致。结论机器学习算法可以单纯通过肝、肾功等血液学检测分类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 a new assist method in the diagnosis of urological carcinoma based on routine hematological examination by deep learning. Method The support vector machine （ SVM）and artificial neural network（ ANN）were applied to distinguish the urological carcinoma by analyzing the data of routine hemato logical examination from 3163 patients including normal persons and urological carcinoma cases. Results The average accuracy rate of deep learning method through 5 cross validation was 92. 05% for urological urological carcinoma cases. The accuracy was no significant differences between SVM and ANN. Conclusions There is probability to classify normal and urological carcinoma patients by using deep learning method on the routine hematological examination.

关 键 词：血液学检测 机器学习 支持向量机 神经网络 泌尿系统肿瘤 疾病筛查 

分 类 号：R737.1[医药卫生—肿瘤]