Research and Development, Aesculap AG, 78532 Tuttlingen, Germany.
Department of Orthopaedic and Trauma Surgery, Musculoskeletal University Center Munich (MUM), Campus Grosshadern, Ludwig Maximilians University Munich, 81377 Munich, Germany.
Sensors (Basel). 2024 May 23;24(11):3324. doi: 10.3390/s24113324.
Recently, inertial measurement units have been gaining popularity as a potential alternative to optical motion capture systems in the analysis of joint kinematics. In a previous study, the accuracy of knee joint angles calculated from inertial data and an extended Kalman filter and smoother algorithm was tested using ground truth data originating from a joint simulator guided by fluoroscopy-based signals. Although high levels of accuracy were achieved, the experimental setup leveraged multiple iterations of the same movement pattern and an absence of soft tissue artefacts. Here, the algorithm is tested against an optical marker-based system in a more challenging setting, with single iterations of a loaded squat cycle simulated on seven cadaveric specimens on a force-controlled knee rig. Prior to the optimisation of local coordinate systems using the REference FRame Alignment MEthod (REFRAME) to account for the effect of differences in local reference frame orientation, root-mean-square errors between the kinematic signals of the inertial and optical systems were as high as 3.8° ± 3.5° for flexion/extension, 20.4° ± 10.0° for abduction/adduction and 8.6° ± 5.7° for external/internal rotation. After REFRAME implementation, however, average root-mean-square errors decreased to 0.9° ± 0.4° and to 1.5° ± 0.7° for abduction/adduction and for external/internal rotation, respectively, with a slight increase to 4.2° ± 3.6° for flexion/extension. While these results demonstrate promising potential in the approach's ability to estimate knee joint angles during a single loaded squat cycle, they highlight the limiting effects that a reduced number of iterations and the lack of a reliable consistent reference pose inflicts on the sensor fusion algorithm's performance. They similarly stress the importance of adapting underlying assumptions and correctly tuning filter parameters to ensure satisfactory performance. More importantly, our findings emphasise the notable impact that properly aligning reference-frame orientations before comparing joint kinematics can have on results and the conclusions derived from them.
最近，惯性测量单元作为一种替代光学运动捕捉系统分析关节运动学的潜在方法，越来越受到关注。在之前的一项研究中，使用基于荧光透视信号引导的关节模拟器生成的真实数据，测试了通过惯性数据和扩展卡尔曼滤波和平滑算法计算的膝关节角度的准确性。尽管达到了高精度，但实验设置利用了相同运动模式的多次迭代和不存在软组织伪影。在这里，该算法在更具挑战性的设置下与基于光学标记的系统进行了测试，在力控制的膝关节装置上对七个尸体标本进行了加载深蹲周期的单次迭代模拟。在使用参考框架对齐方法（REFRAME）优化局部坐标系以考虑局部参考框架方向差异的影响之前，惯性和光学系统的运动信号之间的均方根误差高达 3.8°±3.5°，用于屈伸，20.4°±10.0°用于外展/内收，8.6°±5.7°用于外旋/内旋。然而，在实施 REFRAME 之后，平均均方根误差分别降低到 0.9°±0.4°和 1.5°±0.7°，用于外展/内收和外旋/内旋，而屈伸时略有增加到 4.2°±3.6°。虽然这些结果表明该方法在单次加载深蹲周期中估计膝关节角度的能力具有很大的潜力，但它们突出了迭代次数减少和缺乏可靠一致的参考姿势对传感器融合算法性能的限制影响。它们同样强调了适应底层假设和正确调整滤波器参数以确保满意性能的重要性。更重要的是，我们的发现强调了在比较关节运动学之前正确对齐参考框架方向对结果和从中得出的结论的显著影响。
