人工智能机器学习模型预测关节置换术后效果的进展

doi:10.3877/cma.j.issn.1674-134X.2025.06.013

近年来，机器学习(ML)模型在关节置换术后效果的预测方面取得了显著进展。尤其在预测术后功能康复、并发症等方面的准确性上远超传统的统计学方法。ML模型能够整合多样化的数据源如人口统计学、临床、影像学等，并捕捉复杂的非线性关系，从而提供更精确的个体化风险预测。然而，ML模型应用在临床的道路依然困难重重，包括医疗中心数据标准的缺乏导致整合困难、高性能模型的"黑箱"性质导致的信任问题，以及模型的泛化能力、法律监管和合规性等重要议题，以及模型的泛化能力、法律监管和数据安全性问题等，本综述旨在探讨ML模型在关节置换术后效果预测中的研究进展。

关键词: 关节成形术, 人工智能, 机器学习

Recently, machine learning (ML) models have made significant progress in predicting postoperative outcomes after joint replacement. Particularly in predicting postoperative functional recovery and complications, their accuracy far surpasses that of traditional statistical methods. ML models can integrate diverse data sources (such as demographic, clinical, and imaging data) and capture complex nonlinear relationships, thereby providing more precise individualized risk predictions. However, the clinical application of ML models still faces multiple challenges, including difficulties in data integration due to the lack of standardized medical data, trust issues arising from the "black box" nature of high-performance models, as well as constraints related to model generalization, legal regulations and compliance, and data security. This review aimed to explore the research progress of ML models in predicting postoperative outcomes after joint replacement.

Key words: Arthroplasty, Artificial intelligence, Machine learning

图1 人工智能各分支示意图

Figure 1 Schematic diagram of various branches of artificial intelligence

表1 关节置换领域常用的机器学习算法

Table 1 Common machine learning algorithm in joint replacement field

算法名称	原理	优点	缺点	适用场景
线性回归^[11]	建立自变量和因变量之间的线性关系，找到最佳拟合直线。	简单易懂，解释性强；计算效率高；应用广泛。	对异常值敏感；只能处理线性关系。	预测连续型数值变量，如预测患者术后恢复时间。
逻辑回归^[11]	将线性回归的结果通过sigmoid函数映射到0-1之间，用于分类问题。	简单易解释；适用于二分类问题；计算速度快。	不能直接处理多分类问题；对异常值敏感。	预测二分类问题，如预测患者是否会发生感染。
Lasso回归^[12,13,14]	在线性回归的基础上引入L1正则化，可以进行特征选择。	可以进行特征选择，提高模型的解释性；防止过拟合。	对异常值敏感；计算复杂度高。	具有大量特征的高维数据，需要进行特征选择。
Ridge回归^[12,13,14]	在线性回归的基础上引入L2正则化，可以提高模型的泛化能力。	可以提高模型的泛化能力，防止过拟合。	不能进行特征选择。	存在多重共线性问题的数据集。
决策树^[12]	通过不断划分数据集，生成树状结构，最终得到分类或回归结果。	易于理解和解释；可以处理分类和回归问题；可以处理非线性关系。	容易过拟合；对噪声数据敏感。	探索数据特征，可视化决策过程。
随机森林^[12]	多个决策树的集成，通过投票或取平均值得到最终结果。	准确率高；泛化能力强；可以处理高维数据；可以处理缺失值。	模型复杂，难以解释；计算成本高。	大规模数据集，要求高准确率地分类或回归问题。

表1 关节置换领域常用的机器学习算法

Table 1 Common machine learning algorithm in joint replacement field

算法名称	原理	优点	缺点	适用场景
线性回归^[11]	建立自变量和因变量之间的线性关系，找到最佳拟合直线。	简单易懂，解释性强；计算效率高；应用广泛。	对异常值敏感；只能处理线性关系。	预测连续型数值变量，如预测患者术后恢复时间。
逻辑回归^[11]	将线性回归的结果通过sigmoid函数映射到0-1之间，用于分类问题。	简单易解释；适用于二分类问题；计算速度快。	不能直接处理多分类问题；对异常值敏感。	预测二分类问题，如预测患者是否会发生感染。
Lasso回归^[12,13,14]	在线性回归的基础上引入L1正则化，可以进行特征选择。	可以进行特征选择，提高模型的解释性；防止过拟合。	对异常值敏感；计算复杂度高。	具有大量特征的高维数据，需要进行特征选择。
Ridge回归^[12,13,14]	在线性回归的基础上引入L2正则化，可以提高模型的泛化能力。	可以提高模型的泛化能力，防止过拟合。	不能进行特征选择。	存在多重共线性问题的数据集。
决策树^[12]	通过不断划分数据集，生成树状结构，最终得到分类或回归结果。	易于理解和解释；可以处理分类和回归问题；可以处理非线性关系。	容易过拟合；对噪声数据敏感。	探索数据特征，可视化决策过程。
随机森林^[12]	多个决策树的集成，通过投票或取平均值得到最终结果。	准确率高；泛化能力强；可以处理高维数据；可以处理缺失值。	模型复杂，难以解释；计算成本高。	大规模数据集，要求高准确率地分类或回归问题。

[1]	Gross A, Suh A, Abdelmaksoud A, et al. 292 Long-term clinical outcomes and survival analysis of robotic-assisted versus conventional total knee arthroplasty: an 8-year propensity-matched cohort study[J]. Am J Med Sci, 2025, 369: S189-S190.
[2]	Niculescu V, Nistor-Cseppento DC, Tirla S, et al. A retrospective study on the incidence of periprosthetic fractures related to total hip arthroplasty and postoperative complications during hospitalization[J/OL]. Clin Pract, 2025, 15（3）: 42. DOI: 10.3390/clinpract15030042.
[3]	Niculescu V, Dimitriu AL, Nistor-Cseppento DC, et al. Multicenter study of comorbidities in patients with periprosthetic fractures after total hip arthroplasty and their association with immediate postoperative complications[J/OL]. Clin Pract, 2025, 15（6）: 110. DOI: 10.3390/clinpract15060110.
[4]	Lu F, Li C, Hu P, et al. Effects of personalized 3D-printed blocks in total knee arthroplasty and revision surgery for massive bone defects: a single-center retrospective study[J/OL]. J Int Med Res, 2025, 53（1）: 03000605241308383. DOI: 10.1177/03000605241308383.
[5]	de Geofroy B, Fauchille T, Djian M, et al. Individualized total knee arthroplasty achieves better functional results than off-the-shelf implants in patients undergoing personalized coronal alignment[J]. Knee, 2025, 53: 228-235.
[6]	Carr AJ, Robertsson O, Graves S, et al. Knee replacement[J]. Lancet, 2012, 379（9823）: 1331-1340.
[7]	Hirschmann MT, von Eisenhart-Rothe R, Graichen H, et al. AI may enable robots to make a clinical impact in total knee arthroplasty, where navigation has not![J/OL]. J Exp Orthop, 2024, 11（4）: e70061. DOI: 10.1002/jeo2.70061.
[8]	Andriollo L, Picchi A, Iademarco G, et al. The role of artificial intelligence and emerging technologies in advancing total hip arthroplasty[J/OL]. J Pers Med, 2025, 15（1）: 21. DOI: 10.3390/jpm15010021.
[9]	Lan Q, Li S, Zhang J, et al. Reliable prediction of implant size and axial alignment in AI-based 3D preoperative planning for total knee arthroplasty[J/OL]. Sci Rep, 2024, 14（1）: 16971. DOI: 10.1038/s41598-024-67276-3.
[10]	Karthikeyan A, Priyakumar UD. Artificial intelligence: machine learning for chemical sciences[J/OL]. J Chem Sci, 2021, 134（1）: 2. DOI: 10.1007/s12039-021-01995-2.
[11]	Choi RY, Coyner AS, Kalpathy-Cramer J, et al. Introduction to machine learning, neural networks, and deep learning[J/OL]. Transl Vis Sci Technol, 2020, 9（2）: 14. DOI: 10.1167/tvst.9.2.14.
[12]	Sarker IH. Machine learning: algorithms, real-world applications and research directions[J/OL]. SN Comput Sci, 2021, 2（3）: 160. DOI: 10.1007/s42979-021-00592-x.
[13]	Garland A, Bülow E, Lenguerrand E, et al. Prediction of 90-day mortality after total hip arthroplasty[J]. Bone Joint J, 2021, 103-B（3）: 469-478.
[14]	Xie C, Ou S, Lin Z, et al. Prediction of 90-day local complications in patients after total knee arthroplasty: a nomogram with external validation[J/OL]. Orthop J Sports Med, 2022, 10（2）: 23259671211073331. DOI: 10.1177/23259671211073331.
[15]	Johnson K, Orfanos A, Chen E, et al. Machine learning to predict length of stay following revision hip arthroplasty[J]. J Hip Surg, 2024, 8（2）: 84-89.
[16]	Chong YY, Lau CML, Jiang T, et al. Predicting periprosthetic joint infection in primary total knee arthroplasty: a machine learning model integrating preoperative and perioperative risk factors[J/OL]. BMC Musculoskelet Disord, 2025, 26（1）: 241. DOI: 10.1186/s12891-025-08296-6.
[17]	Cheng Z, Li J, Wu W, et al. Developing and validating a machine learning model to predict chronic pain following total knee arthroplasty[J]. Knee, 2025, 56: 52-65.
[18]	Buddhiraju A, Shimizu MR, Seo HH, et al. Generalizability of machine learning models predicting 30-day unplanned readmission after primary total knee arthroplasty using a nationally representative database[J]. Med Biol Eng Comput, 2024, 62（8）: 2333-2341.
[19]	Hunter J, Soleymani F, Viktor H, et al. Using unsupervised machine learning to predict quality of life after total knee arthroplasty[J]. J Arthroplasty, 2024, 39（3）: 677-682.
[20]	Baxter NB, Lin CH, Wallace BI, et al. Development of a machine learning model to predict the use of surgery in patients with rheumatoid arthritis[J]. Arthritis Care Res, 2024, 76（5）: 636-643.
[21]	Chen TL, Buddhiraju A, Bacevich BM, et al. Predicting 30-day reoperation following primary total knee arthroplasty: machine learning model outperforms the ACS risk calculator[J]. Med Biol Eng Comput, 2025, 63（4）: 1131-1141.
[22]	Lee DW, Han HS, Lee MC, et al. Prediction of postoperative gait speed change after bilateral primary total knee arthroplasty in female patients using a machine learning algorithm[J/OL]. Orthop Traumatol Surg Res, 2024, 110（7）: 103842. DOI: 10.1016/j.otsr.2024.103842.
[23]	Kunze KN, Polce EM, Sadauskas AJ, et al. Development of machine learning algorithms to predict patient dissatisfaction after primary total knee arthroplasty[J]. J Arthroplasty, 2020, 35（11）: 3117-3122.
[24]	Devana SK, Shah AA, Lee C, et al. A novel, potentially universal machine learning algorithm to predict complications in total knee arthroplasty[J]. Arthroplast Today, 2021, 10: 135-143.
[25]	Saiki Y, Kabata T, Ojima T, et al. Machine learning algorithm to predict worsening of flexion range of motion after total knee arthroplasty[J]. Arthroplast Today, 2022, 17: 66-73.
[26]	Klemt C, Cohen-Levy WB, Robinson MG, et al. Can machine learning models predict failure of revision total hip arthroplasty?[J]. Arch Orthop Trauma Surg, 2023, 143（6）: 2805-2812.
[27]	Gabriel RA, Harjai B, Prasad RS, et al. Machine learning approach to predicting persistent opioid use following lower extremity joint arthroplasty[J]. Reg Anesth Pain Med, 2022, 47（5）: 313-319.
[28]	Sniderman J, Stark RB, Schwartz CE, et al. Patient factors that matter in predicting hip arthroplasty outcomes: a machine-learning approach[J]. J Arthroplasty, 2021, 36（6）: 2024-2032.
[29]	Zhang S, Chen JY, Pang HN, et al. Development and internal validation of machine learning algorithms to predict patient satisfaction after total hip arthroplasty[J/OL]. Arthroplasty, 2021, 3（1）: 33. DOI: 10.1186/s42836-021-00087-3.
[30]	El-Othmani MM, Zalikha AK, Shah RP. Comparative analysis of the ability of machine learning models in predicting in-hospital postoperative outcomes after total hip arthroplasty[J]. J Am Acad Orthop Surg, 2022, 30（20）: e1337-e1347.
[31]	Fujii J, Aoyama S, Tezuka T, et al. Prediction of change in pelvic tilt after total hip arthroplasty using machine learning[J]. J Arthroplasty, 2023, 38（10）: 2009-2016. e3.
[32]	Manning DW, Edelstein AI, Alvi HM. Risk prediction tools for hip and knee arthroplasty[J]. J Am Acad Orthop Surg, 2016, 24（1）: 19-27.
[33]	Bozic KJ, Rubash HE, Sculco TP, et al. An analysis of medicare payment policy for total joint arthroplasty[J]. J Arthroplasty, 2008, 23（6 Suppl 1）: 133-138.
[34]	Connolly P, Coombs S, Schwarzkopf R. Mechanical complications after total knee arthroplasty[J]. Expert Rev Med Devices, 2023, 20（12）: 1105-1117.
[35]	Evangelopoulos DS, Ahmad SS, Krismer AM, et al. Periprosthetic infection: major cause of early failure of primary and revision total knee arthroplasty[J]. J Knee Surg, 2019, 32（10）: 941-946.
[36]	Meyer JA, Zhu M, Cavadino A, et al. Infection and periprosthetic fracture are the leading causes of failure after aseptic revision total knee arthroplasty[J]. Arch Orthop Trauma Surg, 2021, 141（8）: 1373-1383.
[37]	Shohat N, Ludwick L, Sherman MB, et al. Using machine learning to predict venous thromboembolism and major bleeding events following total joint arthroplasty[J/OL]. Sci Rep, 2023, 13（1）: 2197. DOI: 10.1038/s41598-022-26032-1.
[38]	Kremers HM, Sohn S, Kremers WK. Machine learning models for prediction of joint infections following hip replacement surgery[C]//2023 IEEE International Conference on Bioinformatics and Biomedicine （BIBM）. December 5-8, 2023, Istanbul, Turkiye. IEEE, 2024: 3053-3058.
[39]	Abraham VM, Booth G, Geiger P, et al. Machine-learning models predict 30-day mortality, cardiovascular complications, and respiratory complications after aseptic revision total joint arthroplasty[J]. Clin Orthop Relat Res, 2022, 480（11）: 2137-2145.
[40]	D’Arrigo G, Gori M, Pitino A, etal. Statistical methods to assess the prognostic value of risk prediction rules in clinical research[J]. Aging Clin Exp Res, 2021, 33（2）: 279-283.
[41]	Pua Y-H, Kang H, Thumboo J, et al. Machine learning methods are comparable to logistic regression techniques in predicting severe walking limitation following total knee arthroplasty[J]. Knee Surg Sports Traumatol Arthrosc, 2020, 28（10）: 3207-3216.
[42]	Ayers DC, Yousef M, Yang W, et al. Age-related differences in pain, function, and quality of life following primary total knee arthroplasty: results from a FORCE-TJR （function and outcomes research for comparative effectiveness in total joint replacement） cohort[J]. J Arthroplasty, 2023, 38（7 Suppl 2）: S169-S176.
[43]	Xu C, Guo H, Wang Q, et al. Interaction of obesity with smoking and inflammatory arthropathies increases the risk of periprosthetic joint infection: a propensity score matched study in a Chinese Han population[J]. J Hosp Infect, 2019, 101（2）: 222-228.
[44]	Pua YH, Seah FJ, Seet FJ, et al. Sex differences and impact of body mass index on the time course of knee range of motion, knee strength, and gait speed after total knee arthroplasty[J]. Arthritis Care Res, 2015, 67（10）: 1397-1405.
[45]	Fontana MA, Lyman S, Sarker GK, et al. Can machine learning algorithms predict which patients will achieve minimally clinically important differences from total joint arthroplasty?[J]. Clin Orthop Relat Res, 2019, 477（6）: 1267-1279.
[46]	Qamar T, Bawany NZ. Understanding the black-box: towards interpretable and reliable deep learning models[J/OL]. PeerJ Comput Sci, 2023, 9: e1629. DOI: 10.7717/peerj-cs.1629.
[47]	Rudin C. Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead[J]. Nat Mach Intell, 2019, 1（5）: 206-215.
[48]	Vela D, Sharp A, Zhang R, et al. Temporal quality degradation in AI models[J/OL]. Sci Rep, 2022, 12: 11654. DOI: 10.1038/s41598-022-15245-z.
[49]	张思平, 刘伟, 马鹏程. 全膝关节置换术后下肢轻度内翻对线对疗效的影响[J/OL]. 中华关节外科杂志（电子版）, 2023, 17（6）: 808-817.
[50]	Colyn W, Bruckers L, Scheys L, et al. Changes in coronal knee-alignment parameters during the osteoarthritis process in the Varus knee[J]. J ISAKOS, 2023, 8（2）: 68-73.

[1]	周欣, 梁豪进, 邓振宇, 肖菊花, 周小军. 基于人工智能技术评价江西省孕11~13⁺⁶周产前超声筛查质量现状及提出能力提升对策[J/OL]. 中华医学超声杂志（电子版）, 2025, 22(09): 850-857.
[2]	张振奇, 齐艺涵, 王璐, 胡紫玥, 李婷婷, 卢漫. 大语言模型DeepSeek-R1在甲状腺超声报告质量控制中的初步应用[J/OL]. 中华医学超声杂志（电子版）, 2025, 22(09): 832-837.
[3]	江瑶, 蒋程, 余翔, 谭莹, 温昕, 温慧莹, 彭桂艳, 李胜利. 基于注意力机制改进的子宫解剖结构检测与分割多任务模型的性能评估[J/OL]. 中华医学超声杂志（电子版）, 2025, 22(08): 703-710.
[4]	陈明朗, 许凯, 黄稚熙, 梁博诚, 贺杰, 黄海珊, 马微波, 谭莹, 邹志英, 刘晓棠, 彭桂艳, 陈家希, 钟晓红. MobileNetV4：面向产前超声的主动脉弓分支异常智能诊断研究[J/OL]. 中华医学超声杂志（电子版）, 2025, 22(08): 711-720.
[5]	杨丽仙, 黄稚熙, 梁博诚, 欧阳淑媛, 陈明朗, 赵英丽, 马薇波, 缪敬, 王磊, 袁鹰. 基于产前时序超声数据的新生儿出生体重智能预测[J/OL]. 中华医学超声杂志（电子版）, 2025, 22(08): 721-732.
[6]	刘晴晴, 俞劲, 徐玮泽, 张志伟, 潘晓华, 舒强, 叶菁菁. OBICnet图像分类模型在小儿先天性心脏病超声筛查中的应用价值[J/OL]. 中华医学超声杂志（电子版）, 2025, 22(08): 754-760.
[7]	傅小芳, 杨青翰, 孙昌琴, 豆梦杰, 胡峻溥, 孙灏, 吕发勤. 基于YOLO 11的肢体长骨骨折断端超声检测模型的临床价值[J/OL]. 中华医学超声杂志（电子版）, 2025, 22(06): 541-546.
[8]	夏堃, 佘蔡楠, 杨昊天, 陶然, 陆跃, 马洪冬. 膝外翻全膝关节置换术后下肢力线与疗效的相关性[J/OL]. 中华关节外科杂志(电子版), 2025, 19(06): 677-683.
[9]	国家骨科医学中心保膝联盟, 中国老年保健协会骨关节保护与健康分会, 单髁置换术后感染抗生素使用专家共识工作组. 单髁置换术后假体周围感染的抗生素使用专家共识（2025版）[J/OL]. 中华损伤与修复杂志(电子版), 2025, 20(06): 476-489.
[10]	杨雯林, 吴元魁. 影像组学在胰腺神经内分泌瘤诊疗中的研究进展[J/OL]. 中华普通外科学文献(电子版), 2025, 19(06): 426-432.
[11]	毛俊, 蔡兆伦, 尹晓南, 沈朝勇, 张波. 影像组学预测模型在胃肠间质瘤诊断及预后中的研究进展[J/OL]. 中华普通外科学文献(电子版), 2025, 19(06): 421-425.
[12]	嵇振岭. 疝与腹壁外科补片研发国外进展[J/OL]. 中华疝和腹壁外科杂志(电子版), 2025, 19(06): 614-622.
[13]	刘宝德, 郑娉娉, 肖宇翔, 沈凌, 王柳盛, 张维. 肺泡灌洗液中炎症因子对重症肺炎患者痰栓风险的预测分析[J/OL]. 中华肺部疾病杂志(电子版), 2025, 18(06): 917-922.
[14]	薛怡宁, 兰雅迪, 刘兆宇, 史磊, 赵琪, 许洪伟. 基于图像的人工智能在胃癌中的研究进展[J/OL]. 中华消化病与影像杂志(电子版), 2025, 15(06): 670-675.
[15]	常芳媛, 乔春梅, 王欣, 王博冉, 赵梓孚, 李春歌, 王晓磊. 多模态超声及人工智能在细菌性和非细菌性关节炎中应用的研究进展[J/OL]. 中华临床医师杂志(电子版), 2025, 19(08): 606-611.

阅读次数

全文

摘要

选择文件类型/文献管理软件名称

选择包含的内容

Progress on artificial intelligence machine learning models in predicting postoperative outcomes in joint replacement

模态框（Modal）标题

选择文件类型/文献管理软件名称

选择包含的内容

Progress on artificial intelligence machine learning models in predicting postoperative outcomes in joint replacement