北京大学学报(医学版) ›› 2019, Vol. 51 ›› Issue (4): 653-659. doi: 10.19723/j.issn.1671-167X.2019.04.010

• 论著 • 上一篇    下一篇

机器学习模型在预测肾结石输尿管软镜碎石术后早期结石清除率中的应用

朱学华1,杨明钰2,夏海缀1,何为1,张智荧1,刘余庆1,(),肖春雷1,(),马潞林1,卢剑1   

  1. 1. 北京大学第三医院泌尿外科,北京 100191
    2. 北京大学信息科学与技术学院,北京 100871
  • 收稿日期:2019-03-14 出版日期:2019-08-18 发布日期:2019-09-03
  • 通讯作者: 刘余庆,肖春雷 E-mail:pku3uro@aliyun.com;xiaochunleixcl@sina.com
  • 基金资助:
    北京大学智慧医疗专项(BMU2018ZHYL012)

Application of machine learning models in predicting early stone-free rate after flexible ureteroscopic lithotripsy for renal stones

Xue-hua ZHU1,Ming-yu YANG2,Hai-zhui XIA1,Wei HE1,Zhi-ying ZHANG1,Yu-qing LIU1,(),Chun-lei XIAO1,(),Lu-lin MA1,Jian LU1   

  1. 1. Department of Urology, Peking University Third Hospital, Beijing 100191, China
    2. School of Electronics Engineering and Computer Science, Peking University, Beijing 100871, China
  • Received:2019-03-14 Online:2019-08-18 Published:2019-09-03
  • Contact: Yu-qing LIU,Chun-lei XIAO E-mail:pku3uro@aliyun.com;xiaochunleixcl@sina.com
  • Supported by:
    Supported by the Intelligent Medical Program of Peking University(BMU2018ZHYL012)

摘要:

目的:基于随机森林和XGBoost两种机器学习算法建立预测模型,探讨其对肾结石患者行输尿管软镜碎石术(flexible ureteroscopic lithotripsy,fURL)后早期结石清除率(stone-free rate, SFR)的预测价值。方法:回顾性分析201例行fURL的肾结石患者的临床资料,根据术后是否达到结石清除标准,将患者分为结石清除组和结石残留组。比较两组患者年龄、体重指数(body mass index,BMI)、结石数目、结石体积、结石密度和肾积水等因素的差异。对于肾下盏结石,需测量肾脏解剖相关指标,包括肾盂漏斗部夹角、肾下盏宽度、肾下盏长度及肾盂肾下盏高度。将上述潜在影响因素分别纳入随机森林和XGBoost算法建立预测模型,绘制受试者工作曲线,检验模型预测价值。前瞻性收集71例患者的临床资料对模型进行外部验证。结果:201例fURL手术均顺利完成,一期手术早期SFR为61.2%。利用随机森林和XGBoost算法建立预测模型并得到不同变量预测重要性评分,随机森林模型和XGBoost模型曲线下面积均为0.77。应用71例样本对模型进行外部验证结果显示,随机森林模型对检测样本的预测总准确率、特异度及灵敏度分别为74.6%、82.6%和60.0%,XGBoost模型对检测样本的预测总准确率、特异度及灵敏度分别为80.3%、87.0%和68.0%。在两种模型中,预测重要性评分排名前四位的变量均为结石体积、平均结石密度、最大结石密度和BMI。结论:基于随机森林和XGBoost算法建立的机器学习模型可准确预测肾结石患者fURL术后早期结石清除状态,有利于术前评估及临床决策。结石体积、平均结石密度、最大结石密度和BMI可能是影响肾结石fURL术后SFR的重要预测因素。

关键词: 机器学习, 随机森林, XGBoost, 肾结石, 结石清除率, 预测模型

Abstract:

Objective: To establish predictive models based on random forest and XGBoost machine learning algorithm and to investigate their value in predicting early stone-free rate (SFR) after flexible ureteroscopic lithotripsy (fURL) in patients with renal stones.Methods: The clinical data of 201 patients with renal stones who underwent fURL were retrospectively investigated. According to the stone-free standard, the patients were divided into stone-free group (SF group) and stone-residual group (SR group). We compared a number of factors including patient age, body mass index (BMI), stone number, stone volume, stone density and hydronephrosis between the two groups. For low calyceal calculi, renal anatomic parameters including infundibular angle (IPA), infundibular width (IW), infundibular length (IL) and pelvic calyceal height (PCH), would be measured. We brought above potential predictive factors into random forest and XGBoost machine learning algorithm respectively to develop two predictive models. The receiver operating characteristic curve (ROC curve) was established in order to test the predictive ability of the model. Clinical data of 71 patients were collected prospectively to validate the predictive models externally.Results: In this study, 201 fURL operations were successfully completed. The one-phase early SFR was 61.2%. We built two predictive models based on random forest and XGBoost machine learning algorithm. The predictive variables’ importance scores were obtained. The area under the ROC curve (AUROC) of the two predictive models for early stone clearance status prediction was 0.77. In the study, 71 test samples were used for external validation. The results showed that the total predictive accuracy, predictive specificity and predictive sensitivity of the random forest and XGBoost models were 75.7%, 82.6%, 60.0%, and 81.4%, 87.0%, 68.0%, respectively. The first four predictive variables in importance were stone volume, mean stone density, maximal stone density and BMI in both random forest and XGBoost predictive models.Conclusion: The predictive models based on random forest and XGBoost machine learning algorithm can predict postoperative early stone status after fURL for renal stones accurately, which will facilitate preoperative evaluation and clinical decision-making. Stone volume, mean stone density, maximal stone density and BMI may be the important predictive factors affecting early SFR after fURL for renal stones.

Key words: Machine learning, Random forest, XGBoost, Renal stones, Stone-free rate, Predictive model

中图分类号: 

  • R692.4

图1

肾脏解剖结构相关指标测量方法示意图"

表1

结石清除组与结石残留组患者基本信息及结石因素比较"

Items Stone-free group (n=123) Stone-residual group (n=78) P value
Male, n (%) 75.0 (61.0) 44.0 (56.4) 0.163a
Age/year, x?±s 51.4±14.4 50.5±13.4 0.655b
BMI/(kg/m2), median (min-max) 25.0 (18.0-33.1) 24.5 (17.7-35.0) 0.475c
History of urologic operation, n (%) 32.0 (26.0) 27.0 (34.6) 0.192a
Hydronephrosis, n (%) 67.0 (54.5) 53.0 (68.0) 0.058a
Low calyceal stones, n (%) 75 (61.0) 51.0 (65.4) 0.887a
Stone number, median (min-max) 1.0 (1.0-5.0) 1.5 (1.0-5.0) 0.000c
Stone volume/mm3, median (min-max) 492.7 (8.37-15 055.4) 1 406.7 (133.2-11 992.0) 0.000c
Mean stone density/Hu, x?±s 548.7±273.7 734.2±299.1 0.000b
Maximal stone density/Hu, median (min-max) 1 049.0 (240.0-2 134.0) 1 266.00 (304.0-1 819.0) 0.000c
Operation time/min, median (min-max) 89.0 (20.0-326.0) 111.0 (22.0-282.0) 0.000c

表2

结石清除组与结石残留组肾下盏解剖因素比较"

Items Stone-free group (n=75) Stone-residual group (n=51) P value
PCH/mm, x?±s 17.0±4.3 21.8±5.3 0.000b
IPA/(°), median (min-max) 44.3 (29.2-71.5) 34.9 (22.3-44.6) 0.000c
IL/mm, x?±s 22.9±3.7 27.8±5.0 0.000b
IW/mm, median (min-max) 7.3 (1.0-22.0) 6.8 (1.0-9.5) 0.076c

图2

随机森林和XGBoost模型中的变量重要性评分"

图3

随机森林和XGBoost模型预测结石清除率的ROC曲线"

表3

随机森林和XGBoost模型的预测价值"

Items Stone-residual Stone-free Total
Random forest model, n
Positive 15 8 23
Negative 10 38 48
Total 25 46 71
XGBoost model, n
Positive 17 6 23
Negative 8 40 48
Total 25 46 71
[1] Zeng G, Mai Z, Xia S , et al. Prevalence of kidney stones in China: an ultrasonography based cross-sectional study[J]. BJU Int, 2017,120(1):109-116.
[2] Sanguedolce F, Bozzini G, Chew B , et al. The evolving role of retrograde intrarenal surgery in the treatment of urolithiasis[J]. Eur Urol Focus, 2017,3(1):46-55.
[3] Berardinelli F, Proietti S, Cindolo L , et al. A prospective multicenter European study on flexible ureterorenoscopy for the management of renal stone[J]. Int Braz J Urol, 2016,42(3):479-486.
[4] 杨波, 胡卫国, 胡浩 , 等. 逆行肾内手术治疗肾结石失败的原因分析及其对策[J]. 北京大学学报(医学版), 2014,46(5):794-797.
[5] Resorlu B, Unsal A, Gulec H , et al. A new scoring system for predicting stone-free rate after retrograde intrarenal surgery: the “resorlu-unsal stone score”[J]. Urology, 2012,80(3):512-518.
[6] Jung JW, Lee BK, Park YH , et al. Modified Seoul National University Renal Stone Complexity score for retrograde intrarenal surgery[J]. Urolithiasis, 2014,42(4):335-340.
[7] Ito H, Sakamaki K, Kawahara T , et al. Development and internal validation of a nomogram for predicting stone-free status after flexible ureteroscopy for renal stones[J]. BJU Int, 2015,115(3):446-451.
[8] Xiao YL, Li D, Chen L , et al. The R.I.R.S. scoring system: An innovative scoring system for predicting stone-free rate following retrograde intrarenal surgery[J]. BMC Urol, 2017,17(1):105.
[9] Breiman L . Random forests[J]. Mach Learn, 2001,45(1):5-32.
[10] Chen T, Guestrin C. XGboost: A scalable tree boosting system. Proceedings of the ACM SIGKDD international conference on knowledge discovery and data mining [C]. San Francisco, California, USA, 2016: 785-794.
[11] Mei X, Wang R, Yang W , et al. Predicting malignancy of pulmonary ground-glass nodules and their invasiveness by random forest[J]. J Thorac Dis, 2018,10(1):458-463.
[12] Taylor RA, Moore CL, Cheung KH , et al. Predicting urinary tract infection in the emergency department with machine learning[J]. PLoS One, 2018,13(3):e0194085.
[13] 刘可, 肖春雷, 刘余庆 , 等. 标准化技术输尿管软镜钬激光碎石术治疗输尿管近端及肾结石:单中心140例报告[J]. 中国微创外科杂志, 2015,15(12):1065-1068, 1087.
[14] Ghani KR, Wolf JS Jr . What is the stone-free rate following flexible ureteroscopy for kidney stones?[J]. Nat Rev Urol, 2015,12(5):281-288.
[15] Ito H, Kawahara T, Terao H , et al. The most reliable preoperative assessment of renal stone burden as a predictor of stone-free status after flexible ureteroscopy with holmium laser lithotripsy: a single-center experience[J]. Urology, 2012,80(3):524-528.
[16] Resorlu B, Oguz U, Resorlu EB , et al. The impact of pelvica-liceal anatomy on the success of retrograde intrarenal surgery in patients with lower pole renal stone[J]. Urology, 2012,79(1):61-66.
[17] 曹文哲, 应俊, 陈广飞 , 等. 基于Logistic回归和随机森林算法的2型糖尿病并发视网膜病变风险预测及对比研究[J]. 中国医疗设备, 2016,31(3):33-38, 69.
doi: 10.3969/j.issn.1674-1633.2016.03.006
[18] Kang SK, Cho KS, Kang DH , et al. Systematic review and meta-analysis to compare success rate of retrograde intrarenal surgery versus percutaneous nephrolithotomy for renal stones >2 cm: An update[J]. Medicine (Baltimore), 2017,96(49):e9119.
[19] Inoue T, Murota T, Okada S , et al. Influence of pelvicaliceal anatomy on stone clearance after flexible ureteroscopy and holmium laser lithotripsy for large renal stones[J]. J Eudourol, 2015,29(9):998-1005.
[20] Sari S, Ozok Hu, Topaloglu H , et al. The association of a number of anatomical factors with the success of retrograde intrarenal surgery in lower calyceal stones[J]. Urol J, 2017,14(4):4008-4014.
[21] 王训师 . XGBoost机器学习模型在缺血性卒中后早期认知损害诊断中的应用研究[D]. 杭州:浙江大学, 2018.
[22] Jessen JP, Honeck P, Knoll T , et al. Flexible ureterorenoscopy for lower pole stones: influence of the collecting system’s anatomy[J]. J Endourol, 2014,28(2):146-151.
[23] Nicodemus KK, Malley JD, Strobl C , et al. The behavior of random forest permutation-based variable importance measures under predictor correlation[J]. BMC Bioinformatics, 2010,11:110.
[1] 吴静依,林瑜,蔺轲,胡永华,孔桂兰. 基于机器学习的重症监护室超长入住时长预测[J]. 北京大学学报(医学版), 2021, 53(6): 1163-1170.
[2] 林瑜,吴静依,蔺轲,胡永华,孔桂兰. 基于集成学习模型预测重症患者再入重症监护病房的风险[J]. 北京大学学报(医学版), 2021, 53(3): 566-572.
[3] 康宁,蒋一航,蒋宇光,吴栗洋,张际青,牛亦农,张军晖. 内镜联合超声与单纯超声引导建立皮肾通道在多镜联合术中的应用[J]. 北京大学学报(医学版), 2020, 52(4): 692-696.
[4] 王明瑞,王起,胡浩,赖金惠,贺永新,熊杰,刘献辉,刘士军,许克新,徐涛. 标准通道经皮肾镜取石术治疗孤立肾肾结石的长期安全性和有效性[J]. 北京大学学报(医学版), 2020, 52(4): 663-666.
[5] 张茗茗,郑迎东,梁宇红. 牙髓根尖周病根管治疗疗效预测模型的建立[J]. 北京大学学报(医学版), 2018, 50(1): 123-130.
[6] 谌诚,张博,韩文科,林健,王刚,张晓春,宋毅,赵峥,张中元,金杰,虞巍. 超声造影技术引导下经皮肾穿刺应用于经皮肾镜取石术的单中心初步经验[J]. 北京大学学报(医学版), 2017, 49(6): 1071-1075.
[7] 王少刚, 余虓. 经皮肾镜碎石术——日间手术新探索[J]. 北京大学学报(医学版), 2017, 49(5): 753-755.
[8] 刘可, 肖春雷, 刘余庆, 郝一昌, 张树栋, 田雨, 马潞林. 输尿管软镜下钬激光憩室颈部切开及碎石治疗微小出口肾盏憩室结石[J]. 北京大学学报(医学版), 2015, 47(4): 618-621.
[9] 王澍, 施永康, 黄晓波, 马凯, 许清泉, 熊六林, 李建兴, 王晓峰. 上尿路结石合并感染的细菌培养及药物敏感性分析[J]. 北京大学学报(医学版), 2014, 46(5): 798-801.
[10] 杨波, 胡卫国, 胡浩, 陈亮, 李建兴, 王晓峰. 逆行肾内手术治疗肾结石失败的原因分析及其对策[J]. 北京大学学报(医学版), 2014, 46(5): 794-797.
[11] 陈亮, 李建兴, 黄晓波, 王晓峰. 一期经皮肾镜手术治疗无发热结石性脓肾术后发生全身炎症反应综合征的危险因素分析[J]. 北京大学学报(医学版), 2014, 46(4): 566-569.
[12] 赫崇军, 秦彩朋, 李建兴, 熊六林, 许清泉, 杨波, 徐涛, 黄晓波, 王晓峰. 肾结石伴积水合并肾盂肿瘤的诊治(附5例报告)[J]. 北京大学学报(医学版), 2014, 46(4): 558-562.
[13] 王姣,罗功唐,牛伟静,宫曼漫,刘璐,周杰,周雪薇,何丽华. 北京市某区医院肾结石危险因素的病例对照研究[J]. 北京大学学报(医学版), 2013, 45(6): 971-974.
[14] 熊六林,叶雄俊,马凯,黄晓波,王晓峰. 无管化24 F通道经皮肾镜治疗肾输尿管上段结石的初步探讨[J]. 北京大学学报(医学版), 2013, 45(4): 575-.
[15] 熊六林, 黄晓波, 叶雄俊, 马凯, 许清泉, 李建兴, 王晓峰. CQS-01超声气压弹道碎石清石系统在经皮肾镜手术中的应用[J]. 北京大学学报(医学版), 2012, 44(4): 575-578.
Viewed
Full text


Abstract

Cited

  Shared   
  Discussed   
[1] 张三. 中文标题测试[J]. 北京大学学报(医学版), 2010, 42(1): 1 -10 .
[2] 赵磊, 王天龙 . 右心室舒张末期容量监测用于肝移植术中容量管理的临床研究[J]. 北京大学学报(医学版), 2009, 41(2): 188 -191 .
[3] 万有, , 韩济生, John E. Pintar. 孤啡肽基因敲除小鼠电针镇痛作用增强[J]. 北京大学学报(医学版), 2009, 41(3): 376 -379 .
[4] 张燕, 韩志慧, 钟延丰, 王盛兰, 李玲玲, 郑丹枫. 骨骼肌活组织检查病理诊断技术的改进及应用[J]. 北京大学学报(医学版), 2009, 41(4): 459 -462 .
[5] 赵奇, 薛世华, 刘志勇, 吴凌云. 同向施压测定自酸蚀与全酸蚀粘接系统粘接强度[J]. 北京大学学报(医学版), 2010, 42(1): 82 -84 .
[6] 林红, 王玉凤, 吴野平. 学校生活技能教育对小学三年级学生行为问题影响的对照研究[J]. 北京大学学报(医学版), 2007, 39(3): 319 -322 .
[7] 丰雷, 程嘉, 王玉凤. 注意缺陷多动障碍儿童的运动协调功能[J]. 北京大学学报(医学版), 2007, 39(3): 333 -336 .
[8] 李岳玲, 钱秋瑾, 王玉凤. 儿童注意缺陷多动障碍成人期预后及其预测因素[J]. 北京大学学报(医学版), 2007, 39(3): 337 -340 .
[9] . 书讯[J]. 北京大学学报(医学版), 2007, 39(3): 225 -328 .
[10] 牟向东, 王广发, 刁小莉, 阙呈立. 肺黏膜相关淋巴组织型边缘区B细胞淋巴瘤一例[J]. 北京大学学报(医学版), 2007, 39(4): 346 -350 .