Departamento de Ingeniería Biomédica, Vicerrectoría de Ciencias de la Salud, Universidad de Monterrey, San Pedro Garza García, Nuevo León, México.
Occupational Biomechanics and Ergonomics Laboratory, Michael Feil and Ted Oberfeld/CRIR Research Centre, Jewish Rehabilitation Hospital, Laval, QC, Canada.
PLoS One. 2018 Nov 26;13(11):e0207945. doi: 10.1371/journal.pone.0207945. eCollection 2018.
Freezing, an episodic movement breakdown that goes from disrupted gait patterns to complete arrest, is a disabling symptom in Parkinson's disease. Several efforts have been made to objectively identify freezing episodes (FEs), although a standardized methodology to discriminate freezing from normal movement is lacking. Novel mathematical approaches that provide information in the temporal and frequency domains, such as the continuous wavelet transform, have demonstrated promising results detecting freezing, although still with limited effectiveness. We aimed to determine whether a computerized algorithm using the continuous wavelet transform based on baseline (i.e. no movement) rather than on amplitude decrease is more effective detecting freezing. Twenty-six individuals with Parkinson's disease performed two trials of a repetitive stepping-in-place task while they were filmed by a video camera and tracked by a motion capture system. The number of FEs and their total duration were determined from a visual inspection of the videos and from three different computed algorithms. Differences in the number and total duration of the FEs between the video inspection and each of the three methods were obtained. The accuracy to identify the time of occurrence of a FE by each method was also calculated. A significant effect of Method was found for the number (p = 0.016) and total duration (p = 0.013) of the FEs, with the method based on baseline being the closest one to the values reported from the videos. Moreover, the same method was the most accurate in detecting the time of occurrence, and the one reaching the highest sensitivity (88.2%). Findings suggest that threshold detection methods based on baseline and movement amplitude decreases capture different characteristics of Parkinsonian gait, with the first one being more effective at detecting FEs. Moreover, robust approaches that consider both time and frequency characteristics are more sensitive in identifying freezing.
冻结，一种间歇性的运动障碍，表现为步态模式中断直至完全停止，是帕金森病的一种致残症状。尽管缺乏区分冻结与正常运动的标准化方法，但人们已经做出了许多努力来客观地识别冻结发作（FE）。一些新的数学方法，如连续小波变换，提供了时间和频率域的信息，已经证明在检测冻结方面具有很有前景的结果，尽管效果仍然有限。我们旨在确定一种基于基线（即无运动）而不是基于幅度减小的连续小波变换的计算机算法是否更有效地检测冻结。26 名帕金森病患者在视频摄像机拍摄和运动捕捉系统跟踪下进行了两次重复原地踏步任务的试验。FE 的数量及其总持续时间是通过视频的目视检查以及三种不同的计算算法来确定的。从视频检查和三种方法中的每一种方法获得 FE 的数量和总持续时间的差异。还计算了每种方法识别 FE 发生时间的准确性。方法对 FE 的数量（p = 0.016）和总持续时间（p = 0.013）有显著影响，基于基线的方法最接近视频报告的值。此外，相同的方法在检测时间方面最准确，并且达到最高灵敏度（88.2%）。研究结果表明，基于基线和运动幅度减小的阈值检测方法捕捉到了帕金森步态的不同特征，前者在检测 FE 方面更有效。此外，同时考虑时间和频率特征的稳健方法在识别冻结方面更敏感。
