Girod Miguel M, Mulford Kellen L, Kaji Elizabeth S, Grove Austin F, Saniei Sami, Ulrich Marisa N, Taunton Michael J, Hannon Charles P, Trousdale Robert T, Perry Kevin I, Wyles Cody C
Orthopedic Surgery Artificial Intelligence Laboratory, Department of Orthopedic Surgery, Mayo Clinic, Rochester, Minnesota.
Orthopedic Surgery Artificial Intelligence Laboratory, Department of Orthopedic Surgery, Mayo Clinic, Rochester, Minnesota; Mayo Clinic Department of Orthopedic Surgery, Mayo Clinic, Rochester, Minnesota.
J Arthroplasty. 2025 May 19. doi: 10.1016/j.arth.2025.05.052.
Most of the focus regarding total knee arthroplasty (TKA) implant positioning and alignment has been centered on the coronal plane. Posterior condylar offset (PCO) and tibial slope (TS) are sagittal parameters that are measured on radiographs, managed intraoperatively, and are crucial to a stable TKA. We sought to compare whether robotic-assisted TKA (raTKA) versus manual TKA (mTKA) are different with regard to achieving a surgeon's preoperative sagittal targets.
We trained a deep learning model based on a U-Net architecture that calculates PCO and TS on lateral knee radiographs. We deployed this model on a consecutive cohort of 280 patients who underwent either mTKA (n = 132) or raTKA (n = 148), with the same medial stabilized knee implant at a tertiary referral center. Measured resection was the technique for mTKA and either calipered kinematic alignment or gap balancing for raTKA.
Mean absolute error between the algorithm and human measurements was 1.3 ± 1.6° for TS and 1.7 ± 1.4 mm for PCO, which was less than the difference between the human annotators (2.0 ± 1.9° and 2.2 ± 2.6 mm, respectively). Mean difference between goal and postoperative TS was less in raTKA than mTKA (0.3 versus 1.3°; P = 0.03). However, the opposite was observed regarding restoration of native PCO, favoring mTKA (-1.7 versus 3.3 mm; P < 0.001). Overall, despite increased diversity in alignment philosophies and proportion of cementless fixation, there was less variability in raTKA postoperative data, suggesting increased precision.
We developed a deep learning algorithm to calculate PCO and TS on lateral knee radiographs. We observed significant differences between raTKA and mTKA in achieving sagittal plane targets, with raTKA being more precise than mTKA. Future studies are warranted to determine whether these differences are clinically relevant.
全膝关节置换术（TKA）植入物的定位和对线大多聚焦于冠状面。后髁偏移（PCO）和胫骨坡度（TS）是矢状面参数，通过X线片测量，术中调整，对TKA的稳定性至关重要。我们旨在比较机器人辅助TKA（raTKA）与手动TKA（mTKA）在实现外科医生术前矢状面目标方面是否存在差异。
我们基于U-Net架构训练了一个深度学习模型，该模型可在膝关节侧位X线片上计算PCO和TS。我们将此模型应用于一家三级转诊中心连续收治的280例接受mTKA（n = 132）或raTKA（n = 148）的患者队列，这些患者均使用相同的内侧稳定型膝关节植入物。mTKA采用测量截骨技术，raTKA采用卡尺运动学对线或间隙平衡技术。
算法与人工测量之间的平均绝对误差，TS为1.3±1.6°，PCO为1.7±1.4 mm，小于人工标注者之间的差异（分别为2.0±1.9°和2.2±2.6 mm）。raTKA中目标与术后TS的平均差异小于mTKA（0.3°对1.3°；P = 0.03）。然而，在恢复自然PCO方面观察到相反的情况，mTKA更具优势（-1.7 mm对3.3 mm；P < 0.001）。总体而言，尽管对线理念的多样性增加且非骨水泥固定的比例增加，但raTKA术后数据的变异性较小，表明精度提高。
我们开发了一种深度学习算法，用于在膝关节侧位X线片上计算PCO和TS。我们观察到raTKA和mTKA在实现矢状面目标方面存在显著差异，raTKA比mTKA更精确。未来有必要进行研究以确定这些差异是否具有临床相关性。
