Health & Rehabilitation Research Institute, Auckland University of Technology, Auckland, New Zealand.
Health & Rehabilitation Research Institute, Auckland University of Technology, Auckland, New Zealand; Centre for Chiropractic Research, New Zealand College of Chiropractic, Auckland, New Zealand; SMI, Department of Health Science and Technology, Aalborg University, Denmark.
J Electromyogr Kinesiol. 2019 Oct;48:103-111. doi: 10.1016/j.jelekin.2019.06.010. Epub 2019 Jun 27.
A significant challenge in surface electromyography (sEMG) is the accurate identification of onsets and offsets of muscle activations. Manual labelling and automatic detection are currently used with varying degrees of reliability, accuracy and time efficiency. Automatic methods still require significant manual input to set the optimal parameters for the detection algorithm. These parameters usually need to be adjusted for each individual, muscle and movement task. We propose a method to automatically identify optimal detection parameters in a minimally supervised way. The proposed method solves an optimisation problem that only requires as input the number of activation bursts in the sEMG in a given time interval. This approach was tested on an extended version of the widely adopted double thresholding algorithm, although the optimisation could be applied to any detection algorithm. sEMG data from 22 healthy participants performing a single (ankle dorsiflexion) and a multi-joint (step on/off) task were used for evaluation. Detection rate, concordance, F score as an average of sensitivity and precision, degree of over detection, and degree of under detection were used as performance metrices. The proposed method improved the performance of the double thresholding algorithm in multi-joint movement and had the same performance in single joint movement with respect to the performance of the double thresholding algorithm with task specific global parameters. Moreover, the proposed method was robust when an error of up to ±10% was introduced in the number of activation bursts in the optimisation phase regardless of the movement. In conclusion, our optimised method has improved the automation of a sEMG detection algorithm which may reduce the time burden associated with current sEMG processing.
表面肌电图 (sEMG) 的一个重大挑战是准确识别肌肉活动的起始和结束。手动标记和自动检测目前具有不同程度的可靠性、准确性和时间效率。自动方法仍然需要大量的手动输入来为检测算法设置最佳参数。这些参数通常需要根据个体、肌肉和运动任务进行调整。我们提出了一种自动识别最佳检测参数的方法，该方法采用最小监督方式。所提出的方法解决了一个优化问题，该问题只需要输入给定时间间隔内 sEMG 中的激活爆发次数。该方法在广泛采用的双阈值算法的扩展版本上进行了测试，尽管可以将优化应用于任何检测算法。使用来自 22 名健康参与者的 sEMG 数据，他们执行了单一（踝背屈）和多关节（踩踏上下）任务。检测率、一致性、F 分数作为灵敏度和精度的平均值、过检测程度和欠检测程度被用作性能指标。与具有特定任务全局参数的双阈值算法相比，所提出的方法提高了多关节运动中双阈值算法的性能，并且在单关节运动中具有相同的性能。此外，无论运动如何，在优化阶段引入激活爆发次数最多±10%的误差时，所提出的方法具有鲁棒性。总之，我们优化的方法提高了 sEMG 检测算法的自动化程度，这可能会减少当前 sEMG 处理相关的时间负担。
