Ng Gabriel, Andrysek Jan
Institute of Biomedical Engineering, University of Toronto, Toronto, ON, M5S 1A1, Canada.
Bloorview Research Institute, Holland Bloorview Kids Rehabilitation Hospital, Toronto, ON, M4G 1R8, Canada.
J Neuroeng Rehabil. 2025 May 12;22(1):109. doi: 10.1186/s12984-025-01638-4.
Gait quality indices, such as the Gillette Gait Index or Gait Profile Score (GPS), can provide clinicians with objective, straightforward measures to quantify gait pathology and monitor changes over time. However, these methods often require motion capture or stationary gait analysis systems, limiting their accessibility. Inertial sensors offer a portable, cost-effective alternative for gait analysis. This study aimed to evaluate a novel hidden Markov model-based similarity measure (HMM-SM) for assessing gait quality directly from gyroscope and accelerometer data captured by inertial sensors.
Walking trials were conducted with 26 lower-limb prosthetic users and 30 able-bodied individuals, using inertial sensors placed at various lower body locations. We computed the HMM-SM score along with other established inertial sensor-based methods, including the Movement Deviation Profile, Dynamic Time Warping, IMU-based Gait Normalcy Index, and Multifeature Gait Score. Spearman correlations with the GPS, a validated measure of gait quality, were assessed, as well as correlations among the inertial sensor methods. Welch's t-tests were used to evaluate the ability to distinguish between prosthetic subgroups.
The HMM-SM and other inertial sensor-based methods demonstrated moderate-to-strong correlations with the GPS (0.49 <|r|< 0.77 for significant correlations). Comparisons between different measures highlighted key similarities and differences, both in correlations and in their ability to differentiate between subgroups. Overall, the pelvis and lower leg sensors achieved significant correlations and outperformed the upper leg sensors, which did not achieve significant correlations with the GPS for any of the signal-based measures.
Results suggest inertial sensors located at the pelvis and lower leg provide valid markers for monitoring overall gait quality, offering the potential to develop nonobtrusive, wearable systems to facilitate long-term monitoring. Such systems could enhance rehabilitation by enabling continuous gait assessment that can be easily integrated in clinical and everyday settings.
步态质量指标，如吉列步态指数或步态轮廓评分（GPS），可以为临床医生提供客观、直接的测量方法，以量化步态病理学并监测随时间的变化。然而，这些方法通常需要动作捕捉或静态步态分析系统，限制了它们的可及性。惯性传感器为步态分析提供了一种便携、经济高效的替代方案。本研究旨在评估一种基于隐马尔可夫模型的新型相似性度量（HMM-SM），用于直接从惯性传感器捕获的陀螺仪和加速度计数据评估步态质量。
对26名下肢假肢使用者和30名健全个体进行步行试验，在身体下部的不同位置放置惯性传感器。我们计算了HMM-SM评分以及其他基于惯性传感器的既定方法，包括运动偏差轮廓、动态时间规整、基于惯性测量单元的步态正常指数和多特征步态评分。评估了与GPS（一种经过验证的步态质量测量方法）的斯皮尔曼相关性，以及惯性传感器方法之间的相关性。使用韦尔奇t检验评估区分假肢亚组的能力。
HMM-SM和其他基于惯性传感器的方法与GPS显示出中度至强相关性（显著相关性时0.49 < |r| < 0.77）。不同测量方法之间的比较突出了在相关性及其区分亚组能力方面的关键异同。总体而言，骨盆和小腿传感器显示出显著相关性，并且优于大腿传感器，大腿传感器对于任何基于信号的测量方法均未与GPS显示出显著相关性。
结果表明，位于骨盆和小腿的惯性传感器为监测整体步态质量提供了有效的标志物，具有开发无创、可穿戴系统以促进长期监测的潜力。此类系统可以通过实现易于整合到临床和日常环境中的连续步态评估来加强康复治疗。
