| 摘要: |
| ［摘要］ 目的 评估AIJOINT系统预测全膝关节置换术（TKA）假体型号的准确性，并分析影响因素。方法 回顾性分析2025年1月至12月于中山大学附属第一医院关节外科初次接受TKA的412例患者的临床资料，术前采用AIJOINT系统进行人工智能（AI）规划。主要结局指标为预测准确性（精确匹配或±1号内），采用Cohen加权Kappa（κw）评估一致性，bootstrap法计算95%CI。分别以股骨和胫骨假体型号是否精确匹配为因变量，将单因素分析中P<0.10的自变量纳入多因素logistic回归模型，采用后退法筛选变量。结果 股骨假体精确匹配预测准确率为72.8%（300/412），κw=0.752（95%CI：0.694～0.805）；胫骨假体精确匹配预测准确率为70.1%（289/412），κw=0.721（95%CI：0.662～0.773）。±1号内预测准确率股骨为92.0%、胫骨为91.0%。多因素分析结果显示，轻度畸形［OR（95%CI）=8.428（4.499～15.789），P<0.001］、中度畸形［OR（95%CI）=3.106（1.746～5.525），P<0.001］和女性［OR（95%CI）=2.581（1.586～4.202），P<0.001］是股骨假体精确匹配预测准确性的独立影响因素；轻度畸形［OR（95%CI）=2.797（1.597～4.899），P<0.001］、中度畸形［OR（95%CI）=1.898（1.082～3.328），P=0.025］和女性［OR（95%CI）=1.635（1.043～2.564），P=0.032］是胫骨假体精确匹配预测准确性的独立影响因素。结论 AI术前规划预测中国人TKA假体型号具有较高的准确性，畸形程度及性别是预测准确性的独立影响因素，严重畸形患者的预测准确性降低，临床应用时应更谨慎参考AI预测结果。 |
| 关键词: 人工智能 全膝关节置换术 假体型号 术前规划 预测准确性 影响因素 |
| DOI：10.3969/j.issn.1674-3806.2026.05.06 |
| 分类号:R 684 |
| 基金项目:国家重点研发计划子课题（编号：2022YFC2407505） |
|
| Study on the accuracy of AI-based preoperative planning system in predicting component sizes for total knee arthroplasty:a retrospective analysis of 412 cases using AIJOINT system |
|
KANG Yunze1, MAO Guping2, PAN Baiqi1, WU Xiaoyu1, YAO Zeyang1, TU Yucheng1, HAN Tieling1, SHENG Puyi1, ZHANG Ziji1, LI Zhiwen1
|
|
1.Department of Joint Surgery, the First Affiliated Hospital, Sun Yat-sen University, Guangzhou 510080, China; 2.Department of Sports Medicine, the First Affiliated Hospital, Sun Yat-sen University, Guangzhou 510080, China
|
| Abstract: |
| ［Abstract］ Objective To evaluate the accuracy of AIJOINT system in predicting component sizes for total knee arthroplasty(TKA) and to analyze the influencing factors. Methods The clinical data of 412 patients undergoing primary TKA in Department of Joint Surgery, the First Affiliated Hospital, Sun Yat-sen University from January 2025 to December 2025 were retrospectively analyzed, and artificial intelligence(AI)-based preoperative planning was performed using the AIJOINT system.The primary outcome was prediction accuracy(exact match or within ±1 size). The agreement was assessed using Cohen′s weighted Kappa(κw), with 95% confidence interval(CI) calculated by bootstrap resampling. Multivariate logistic regression models were constructed separately for femoral and tibial components whether their sizes were in exact matching as dependent variables, with independent variables showing P<0.10 in univariate analysis entered into the models. The variables were screened by backward stepwise selection. Results The exact matching prediction accuracy rate was 72.8%(300/412) for femoral components［κw=0.752(95%CI: 0.694-0.805)］ and 70.1%(289/412) for tibial components［κw=0.721(95%CI: 0.662-0.773)］. The prediction accuracy for the components within ±1 size was 92.0% for femoral components and 91.0% for tibial components. The results of multivariate analysis showed that mild deformity［OR(95%CI)=8.428(4.499-15.789), P<0.001］, moderate deformity［OR(95%CI)=3.106(1.746-5.525), P<0.001］ and female sex［OR(95%CI)=2.581(1.586-4.202), P<0.001］ were independent influencing factors of exact matching prediction accuracy for femoral components, and mild deformity［OR(95%CI)=2.797(1.597-4.899), P<0.001］, moderate deformity［OR(95%CI)=1.898(1.082-3.328), P=0.025］ and female sex［OR(95%CI)=1.635(1.043-2.564), P=0.032］ were independent influencing factors of exact matching prediction accuracy for tibial components. Conclusion AI-based preoperative planning demonstrates high accuracy in predicting component sizes for TKA in Chinese populations. Deformity severity and gender are independent influencing factors of prediction accuracy. Prediction accuracy is notably reduced in patients with severe deformity, indicating more cautious interpretation of AI predictions in clinical application. |
| Key words: Artificial intelligence(AI) Total knee arthroplasty(TKA) Component sizes Preoperative planning Prediction accuracy Influencing factor |
