切换至 "中华医学电子期刊资源库"

中华医学超声杂志(电子版) ›› 2025, Vol. 22 ›› Issue (02) : 153 -161. doi: 10.3877/cma.j.issn.1672-6448.2025.02.009

介入超声影像学

基于LightGBM 算法构建超声引导下凝血酶治疗假性动脉瘤剂量预测模型
王昕玥1, 袁亚1, 束华1, 曹琨芃1, 叶新华1, 李璐1,()   
  1. 1. 210029 南京医科大学第一附属医院超声诊断科
  • 收稿日期:2024-12-30 出版日期:2025-02-01
  • 通信作者: 李璐

Development of a dose prediction model for ultrasound-guided thrombin treatment of pseudoaneurysms using the LightGBM algorithm

Xinyue Wang1, Ya Yuan1, Hua Shu1, Kunpeng Cao1, Xinhua Ye1, Lu Li1,()   

  1. 1. Department of Ultrasound, the First Affiliated Hospital, Nanjing Medical University, Nanjing 210029, China
  • Received:2024-12-30 Published:2025-02-01
  • Corresponding author: Lu Li
引用本文:

王昕玥, 袁亚, 束华, 曹琨芃, 叶新华, 李璐. 基于LightGBM 算法构建超声引导下凝血酶治疗假性动脉瘤剂量预测模型[J/OL]. 中华医学超声杂志(电子版), 2025, 22(02): 153-161.

Xinyue Wang, Ya Yuan, Hua Shu, Kunpeng Cao, Xinhua Ye, Lu Li. Development of a dose prediction model for ultrasound-guided thrombin treatment of pseudoaneurysms using the LightGBM algorithm[J/OL]. Chinese Journal of Medical Ultrasound (Electronic Edition), 2025, 22(02): 153-161.

目的

基于LightGBM(Light Gradient Boosting Machine)算法,构建超声引导下注射凝血酶治疗假性动脉瘤的剂量预测模型。

方法

回顾性分析2018 年1 月至2024 年12 月期间,在南京医科大学第一附属医院接受超声检查并确诊为股动脉假性动脉瘤且接受超声引导下凝血酶注射治疗的患者84 例。根据凝血酶使用的不同剂量,将患者分为低剂量组(<500 IU)30 例、中剂量组(≥500 IU 且<1000 IU)36 例和高剂量组(≥1000 IU)18 例。按8:2 的比例随机将患者分为训练集67 例,验证集17 例。基于有序Logistic 回归分析筛选特征变量,建立Logistic 凝血酶剂量预测模型。基于LightGBM 算法的贡献度筛选特征变量,构建LightGBM 凝血酶剂量预测模型。模型性能通过总体准确率、微平均曲线下面积(AUC)、召回率、F1 分数等指标进行评估,并通过ROC曲线进行模型效能分析。

结果

基于Logistic 算法的凝血酶剂量预测模型在训练集中的微平均准确率为0.677,微平均AUC 为0.744(95% CI:0.674 ~0.815);在验证集中的微平均准确率为0.686,微平均AUC 为0.758(95%CI:0.624 ~0.891)。基于LightGBM 算法构建的凝血酶剂量预测模型在训练集中的微平均准确率为0.930,微平均AUC 为0.975(95%CI:0.955 ~0.995),微平均召回率为92.5%,微平均F1 分数为0.899;在验证集中的微平均准确率为0.804,微平均AUC 为0.872(95%CI:0.766 ~0.978),微平均召回率为76.5%,微平均F1 分数为0.722。

结论

基于LightGBM算法构建的凝血酶剂量预测模型能够有效预测凝血酶的使用剂量,为实现超声引导下凝血酶注射治疗的精准化和个体化治疗提供了重要参考。

Objective

To develop a dose prediction model for ultrasound-guided thrombin treatment of pseudoaneurysms using the Light Gradient Boosting Machine (LightGBM) algorithm.

Methods

A retrospective analysis was conducted on 84 patients diagnosed with femoral artery pseudoaneurysms via ultrasound and treated with ultrasound-guided thrombin injection at the First Affiliated Hospital of Nanjing Medical University between January 2018 and December 2024.Patients were categorized into three groups based on thrombin dosage:low-dose (<500 IU, 30 cases), mediumdose (≥500 IU and <1000 IU, 36 cases), and high-dose (≥1000 IU, 18 cases).The cohort was randomly divided into a training set (67 cases) and a validation set (17 cases) at an 8:2 ratio.Feature variables were screened using ordinal logistic regression analysis to construct a logistic-based thrombin dose prediction model.Additionally, LightGBM contribution-based feature selection was applied to build a LightGBM-based dose prediction model.Model performance was evaluated using overall accuracy,micro-average area under the curve (AUC), recall, F1-score, and receiver operating characteristic(ROC) curve analysis.

Results

In the training set, the logistic regression model demonstrated an overall accuracy of 0.677 and a micro-average AUC of 0.744 (95% confidence interval [CI]:0.674-0.815); in the validation set, the corresponding values were 0.686 and 0.758 (95%CI:0.624-0.891).The LightGBM-based model exhibited superior performance, with a training set overall accuracy of 0.930,micro-average AUC of 0.975 (95%CI:0.955-0.995), micro-average recall of 92.5%, and micro-average F1-score of 0.899.In the validation set, it achieved an overall accuracy of 0.804, micro-average AUC of 0.872 (95%CI:0.766-0.978), micro-average recall of 76.5%, and micro-average F1-score of 0.722.

Conclusion

The LightGBM-based thrombin dose prediction model effectively forecasts thrombin dosage requirements, offering a valuable reference for achieving precision and individualized treatment in ultrasound-guided thrombin injection therapy.

图1 患者入组流程及研究路线图 注:UGTI 为超声引导下注射凝血酶治疗假性动脉瘤
图2 股动脉假性动脉瘤超声测量示意图。图a 为模式图;图b 为二维超声图像 注:a 为假性动脉瘤瘤腔长径;b 为假性动脉瘤瘤腔短径;c 为瘘管长度;d 为股动脉破口内径
图3 假性动脉瘤超声引导下注射凝血酶治疗术前及术后超声声像图。图a 为术前假性动脉瘤二维超声声像图;图b 为术前假性动脉瘤彩色多普勒血流图;图c 为术后瘤体内血栓形成二维超声声像图;图d 为术后瘤体内血栓形成彩色多普勒血流图
表1 假性动脉瘤患者训练集不同凝血酶剂量的3 组间临床资料比较
临床资料 低剂量(n=24) 中剂量(n=26) 高剂量(n=17) 统计值 P
瘤腔长径(mm, ±s) 21.89±8.69 27.83±9.09 37.17±11.13 F=12.86 < 0.001
瘤腔短径(mm, ±s) 13.06±5.94 15.85±4.77 18.42±5.24 F=5.11 0.009
浓度[IU/ml,MQ1Q3)] 500.00(400.00,541.67) 500.00(500.00,500.00) 500.00(500.00,500.00) Z=2.33 0.312
年龄[ 岁,MQ1Q3)] 62.00(56.00,68.00) 60.50(55.50,67.00) 63.00(58.00,72.00) Z=1.05 0.591
瘘管长度[mm,MQ1Q3)] 4.00(2.08,10.46) 2.55(2.02,4.90) 2.90(1.70,7.40) Z=3.93 0.140
破口内径[mm,MQ1Q3)] 2.10(1.65,2.85) 2.10(1.64,2.80) 2.20(1.80,3.00) Z=0.54 0.762
PSV[cm/s,MQ1Q3)] 170.95(118.75,300.00) 195.00(109.00,296.27) 201.10(170.00,250.00) Z=0.27 0.873
PT[s,MQ1Q3)] 11.80(11.67,16.18) 12.65(11.60,14.80) 12.70(11.70,13.90) Z=0.03 0.984
INR[MQ1Q3)] 1.02(1.01,1.31) 1.10(1.01,1.31) 1.11(1.02,1.22) Z=0.42 0.809
APTT[s,MQ1Q3)] 30.20(27.65,35.38) 30.20(27.27,34.70) 29.40(27.80,33.20) Z=0.04 0.981
FIB[g/L,MQ1Q3)] 2.56(2.11,3.53) 2.44(2.15,3.04) 2.83(2.38,3.08) Z=1.28 0.527
TT[s,MQ1Q3)] 17.25(16.75,17.72) 17.30(16.65,18.05) 17.90(17.30,18.30) Z=4.54 0.103
DD2[mg/L,MQ1Q3)] 0.27(0.15,0.47) 0.33(0.13,1.18) 0.19(0.12,0.24) Z=2.69 0.261
性别[ 例(%)] χ 2=0.62 0.735
14(58.33) 15(57.69) 8(47.06)
10(41.67) 11(42.31) 9(52.94)
抗凝药[ 例(%)] - 0.761
未使用 11(45.83) 13(50.00) 6(35.29)
使用1 种 12(50.00) 13(50.00) 10(58.82)
使用2 种 1(4.17) 0(0.00) 1(5.88)
抗血小板药[ 例(%)] - 0.996
未使用 12(50.00) 14(53.85) 9(52.94)
使用1 种 7(29.17) 7(26.92) 4(23.53)
使用2 种 5(20.83) 5(19.23) 4(23.53)
高血压[ 例(%)] χ 2=0.31 0.855
10(41.67) 10(38.46) 8(47.06)
14(58.33) 16(61.54) 9(52.94)
表2 假性动脉瘤患者凝血酶使用剂量影响因素的有序Logistic 回归分析
表3 Logistic 凝血酶剂量预测模型在训练集和验证集的诊断性能比较
图4 Logistic 凝血酶剂量预测模型的宏平均、微平均及各子类训练集和验证集的ROC 曲线。图a 为训练集;图b 为验证集 注:AUC 为曲线下面积;macro 为宏平均;micro 为微平均;low 为低剂量组;median 为中剂量组;high 为高剂量组
表4 LightGBM 凝血酶剂量预测模型在训练集和验证集的诊断性能比较
图5 LightGBM 凝血酶剂量预测模型的宏平均、微平均及各子类训练集和验证集的ROC 曲线。图a 为训练集;图b 为验证集 注:AUC 为曲线下面积;macro 为宏平均;micro 为微平均;low 为低剂量组;median 为中剂量组;high 为高剂量组
1
Stolt M, Braun-Dullaeus R, Herold J.Do not underestimate the femoral pseudoaneurysm [J].Vasa, 2018, 47(3):177-185.
2
Wischmann P, Stern M, Baasen S, et al.Importance of pseudoaneurysms after TAVI - a retrospective analysis of 2063 patients[J].Vasa, 2025, 54(1):50-58.
3
叶新华, 李璐, 杭菁, 等.超声引导下假性动脉瘤首次凝血酶注射治疗失败1 例原因分析 (附视频) [J].中国临床案例成果数据库,2023, (1):E01717.
4
Peters S, Braun-Dullaeus R, Herold J.Pseudoaneurysm [J].Hamostaseologie, 2018, 38(3):166-172.
5
Rachakonda A, Qato K, Khaddash T, et al.Ultrasound-guided thrombin injection of genicular artery pseudoaneurysm [J].Ann Vasc Surg, 2015, 29(5):1017.e1011-1013.
6
Li L, Deng H, Chen W, et al.Comparison of the diagnostic effectiveness of ultrasound imaging coupled with three mathematical models for discriminating thyroid nodules [J].Acta Radiol, 2024,65(5):441-448.
7
Bafaloukou M, Schalkamp AK, Fletcher-Lloyd N, et al.An interpretable machine learning tool for in-home monitoring of agitation episodes in people living with dementia:a proof-of-concept study [J].EClinicalMedicine, 2025, 80:103032.
8
Li L, Deng H, Yuan Y, et al.Thrombin injection under B-flow and ultrasound guidance:A safe and effective treatment of pseudoaneurysms [J].Vascular, 2024, 32(1):147-153.
9
刘淼, 钱朝庆.超声引导下股动脉假性动脉瘤凝血酶注射和血管腔内缝合临床对比 [J].中国超声医学杂志, 2024, 40(11):1304-1307.
10
Stojanovski BM, Mohammed BM, Di Cera E.The prothrombinprothrombinase interaction [J].Subcell Biochem, 2024, 104:409-423.
11
Bar Barroeta A, Albanese P, Kadavá T, et al.Thrombin activation of the factor XI dimer is a multistaged process for each subunit [J].J Thromb Haemost, 2024, 22(5):1336-1346.
12
Zhao B, Zhang J, Ma J, et al.Comparison of three different treatment methods for traumatic and Iatrogenic peripheral artery pseudoaneurysms [J].Orthop Surg, 2022, 14(7):1404-1412.
13
Mansouri MH, Mansouri P, Hashemi M, et al.Compare efficacy and safety of autologous blood clot injection with C-clamp vascular closure device in treatment of iatrogenic pseudoaneurysm after femoral artery puncture [J].J Vasc Access, 2024:11297298241273641.
14
Müller A, Wouters EF, Koul P, et al.Association between lung function and dyspnoea and its variation in the multinational Burden of Obstructive Lung Disease (BOLD) study [J].Pulmonology, 2025,31(1):2416815.
15
Hu Y, Ma F, Hu M, et al.Development and validation of a machine learning model to predict the risk of readmission within one year in HFpEF patients:Short title:Prediction of HFpEF readmission [J].Int J Med Inform, 2025, 194:105703.
16
Wang L, Wu H, Wu C, et al.A deep-learning system integrating electrocardiograms and laboratory indicators for diagnosing acute aortic dissection and acute myocardial infarction [J].Int J Cardiol,2025, 423:133008.
17
Sun X.Application of an improved LightGBM hybrid integration model combining gradient harmonization and Jacobian regularization for breast cancer diagnosis [J].Sci Rep, 2025, 15(1):2569.
18
Liu Y, Zhang Z, Song H, et al.An improved stacking model for predicting myocardial infarction risk in imbalanced data [J].Health Inf Sci Syst, 2025, 13(1):16.
19
Kleczynski P, Rakowski T, Dziewierz A, et al.Ultrasoundguided thrombin injection in the treatment of iatrogenic arterial pseudoaneurysms:single-center experience [J].J Clin Ultrasound,2014, 42(1):24-26.
20
Mohammad F, Kabbani L, Lin J, et al.Post-procedural pseudoaneurysms:Single-center experience [J].Vascular, 2017, 25(2):178-183.
21
Shin JH, Song Y, Sheen JJ, et al.Safety and effectiveness of percutaneous low-dose thrombin injection for femoral puncture site pseudoaneurysms in neurointervention:single-center experience [J].Neurointervention, 2020, 15(1):25-30.
22
Bortolini E, Leite TFO, Linard B, et al.Ultrasound-guided thrombin injection for cardiac catheterization pseudoaneurysms:efficacy, safety,and predictors [J].Acta Radiol, 2024:2841851241292516.
23
Olsen DM, Rodriguez JA, Vranic M, et al.A prospective study of ultrasound scan-guided thrombin injection of femoral pseudoaneurysm:a trend toward minimal medication [J].J Vasc Surg, 2002, 36(4):779-782.
[1] 洪玮, 叶细容, 刘枝红, 杨银凤, 吕志红. 超声影像组学联合临床病理特征预测乳腺癌新辅助化疗完全病理缓解的价值[J/OL]. 中华医学超声杂志(电子版), 2024, 21(06): 571-579.
[2] 奚玲, 仝瀚文, 缪骥, 毛永欢, 沈晓菲, 杜峻峰, 刘晔. 基于肌少症构建的造口旁疝危险因素预测模型[J/OL]. 中华普外科手术学杂志(电子版), 2025, 19(01): 48-51.
[3] 屈勤芳, 束方莲. 盆腔器官脱垂患者盆底重建手术后压力性尿失禁发生的影响因素及列线图预测模型构建[J/OL]. 中华腔镜泌尿外科杂志(电子版), 2024, 18(06): 606-612.
[4] 李鑫, 李乐, 陈金明, 刘蕊. 基于Meta 分析构建肝癌微血管侵犯的术前预测模型[J/OL]. 中华肝脏外科手术学电子杂志, 2025, 14(01): 74-80.
[5] 公宇, 廖媛, 尚梅. 肝细胞癌TACE术后复发影响因素及预测模型建立[J/OL]. 中华肝脏外科手术学电子杂志, 2024, 13(06): 818-824.
[6] 关小玲, 周文营, 陈洪平. PTAAR在乙肝相关慢加急性肝衰竭患者短期预后中的预测价值[J/OL]. 中华肝脏外科手术学电子杂志, 2024, 13(06): 841-845.
[7] 伍细蓉, 徐立文, 陈亚琼. 基于LPR和FARI构建肝衰竭患者生存预后模型[J/OL]. 中华肝脏外科手术学电子杂志, 2024, 13(05): 675-681.
[8] 王贝贝, 崔振义, 王静, 王晗妍, 吕红芝, 李秀婷. 老年股骨粗隆间骨折患者术后贫血预测模型的构建与验证[J/OL]. 中华老年骨科与康复电子杂志, 2024, 10(06): 355-362.
[9] 任船, 葛荣忠, 邱正才, 朱莉. 肠癌患者术后代谢性酸中毒的预测模型构建及应用价值评价[J/OL]. 中华消化病与影像杂志(电子版), 2025, 15(01): 34-39.
[10] 孙晗, 于冰, 武侠, 周熙朗. 基于循环肿瘤DNA 甲基化的结直肠癌筛查预测模型的构建与验证[J/OL]. 中华消化病与影像杂志(电子版), 2024, 14(06): 500-506.
[11] 韦巧玲, 黄妍, 赵昌, 宋庆峰, 陈祖毅, 黄莹, 蒙嫦, 黄靖. 肝癌微波消融术后中重度疼痛风险预测列线图模型构建及验证[J/OL]. 中华临床医师杂志(电子版), 2024, 18(08): 715-721.
[12] 蔡晓雯, 李慧景, 丘婕, 杨翼帆, 吴素贤, 林玉彤, 何秋娜. 肝癌患者肝动脉化疗栓塞术后疼痛风险预测模型的构建及验证[J/OL]. 中华临床医师杂志(电子版), 2024, 18(08): 722-728.
[13] 孙铭远, 褚恒, 徐海滨, 张哲. 人工智能应用于多发性肺结节诊断的研究进展[J/OL]. 中华临床医师杂志(电子版), 2024, 18(08): 785-790.
[14] 董晟, 郎胜坤, 葛新, 孙少君, 薛明宇. 反向休克指数乘以格拉斯哥昏迷评分对老年严重创伤患者发生急性创伤性凝血功能障碍的预测价值[J/OL]. 中华临床医师杂志(电子版), 2024, 18(06): 541-547.
[15] 洪凡, 王志坚. 瘢痕子宫再妊娠并发症预测模型的研究进展[J/OL]. 中华产科急救电子杂志, 2025, 14(01): 20-25.
阅读次数
全文


摘要