手术相关膝关节表型:基于深度学习的改良CPAK分类系统
本研究基于972例患者(1944膝)的全下肢X线影像,采用深度学习模型自动测量股骨远端外侧角与胫骨近端内侧角,按两层分类框架定义膝关节表型。共识别出7种表型模式,其中5种占99.2%;内翻表型以胫骨驱动为主(99.2%),外翻表型以股骨驱动为主(98.9%)。改良CPAK系统为基于对线的术前规划提供了简化且精确的评估框架。
要点
- 改良CPAK将膝关节分为7种表型,5种主要表型覆盖99.2%的膝关节。
- 内翻膝以胫骨驱动为主(99.2%),外翻膝以股骨驱动为主(98.9%)。
- 中性膝中95%表现为股骨外翻/胫骨内翻组合,完全中性组合不足2%。
核心结论
- 改良CPAK系统为膝关节表型分析提供了简化且精确的框架,支持基于对线的外科规划。
- 膝关节对线呈现一致模式:内翻主要由胫骨驱动,外翻主要由股骨驱动。
研究设计与样本
- 研究纳入来自大型公开数据集的972例患者(1944个健康与关节炎膝关节)的全下肢X线片。
- 采用经验证的深度学习模型自动测量股骨远端外侧角和胫骨近端内侧角。
- 通过两层分类框架确定膝关节表型,同时捕捉整体下肢对线(算术髋-膝-踝角)及股骨与胫骨关节线贡献。
主要发现
- 共识别出7种表型模式,其中5种占膝关节的99.2%。
- 最常见表型为内翻膝伴股骨外翻/胫骨内翻(26.9%),其次为中性膝(25.6%)和外翻膝伴股骨外翻/胫骨内翻(24.2%)。
- 较少见表型包括外翻膝伴股骨外翻/胫骨外翻(12.2%)和内翻膝伴股骨内翻/胫骨内翻(10.3%)。
- 内翻表型以胫骨驱动为主(99.2%),外翻表型以股骨驱动为主(98.9%)。
- 中性膝中95%表现为股骨外翻/胫骨内翻组合,完全中性组合占全队列不足2%。
关键数值与推荐
- 表型分类基于算术髋-膝-踝角评估整体对线,以及股骨远端外侧角和胫骨近端内侧角评估股骨与胫骨关节线贡献。
- 7种表型模式中,5种主要模式覆盖99.2%的膝关节。
- 内翻表型中胫骨驱动占99.2%,外翻表型中股骨驱动占98.9%。
临床怎么用
- 改良CPAK系统可用于基于对线的膝关节手术规划,辅助判断内翻或外翻畸形的来源。
- 内翻膝多源于胫骨,外翻膝多源于股骨,术中可据此选择针对性矫正策略。
以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。
原文
BACKGROUND: The coronal plane alignment of the knee (CPAK) classification proposes nine knee phenotypes based on constitutional limb alignment and joint line obliquity. However, there is considerable confusion and variability in defining types, and its impact on surgical decision-making is unclear. A simple modified CPAK system is proposed to improve accuracy and facilitate decision-making. Herein, we applied a deep learning model to automate knee phenotyping and analyzed modified CPAK distributions in a large cohort. METHODS: We included 972 patients (1,944 healthy and arthritic knees) who had full-limb radiographs from a large publicly available dataset. A previously validated deep learning model was used to measure the lateral distal femoral angle and the medial proximal tibial angle. Knee phenotypes were determined using a two-level classification framework designed to capture both overall limb alignment, as measured by arithmetic hip-knee-ankle, and femoral and tibial joint-line contribution, as measured by lateral distal femoral angle and medial proximal tibial angle. RESULTS: There were seven phenotypic patterns identified, with five accounting for 99.2% of knees. The most frequent phenotypes were varus knees with valgus femur/varus tibia (26.9%), followed by neutral knees (25.6%), and valgus knees with valgus femur/varus tibia (24.2%). Less common were valgus knees with valgus femur/valgus tibia (12.2%) and varus femur/varus tibia in varus knees (10.3%). Varus phenotypes were predominantly tibial-driven (99.2%), while valgus phenotypes were femoral-driven (98.9%). Among neutral knees, 95% demonstrated a valgus femur with varus tibia configuration, while perfectly neutral combinations accounted for fewer than 2% of the total cohort. CONCLUSIONS: The Modifed CPAK offers a simplified and precise framework for knee phenotyping. Knee alignment demostrates a consistent pattern, with varus primarly driven by the tibia and valgus by the femur. This system provides a surgically relevant assessment of the knee, supporting alignment-based surgical planning.
引用
- Burgio C, Zepeda KE, Karasavvidis T, et al. Surgically Relevant Knee Phenotypes: The Modified Coronal Plane Alignment of the Knee System-A Deep Learning-Based Classification. J Arthroplasty. 2026. PMID: 41967528. doi:10.1016/j.arth.2026.04.010
供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。