HD-EMG signals and corresponding neural drive signals for training and validation of a deep CNN for neural drive estimation across muscles and participants

HD-EMG signals and corresponding neural drive signals, in form of composite spike train (CST), for training and validation of a deep CNN framework for neural drive estimation. We included processed data from five participants (uc4, uc8, uc11, uc25, uc26), each with two muscles [vastus lateralis (VL) and the vastus medialis (VM) ] across two sessions (d1 and d2). In each session, they performed three isometric contractions at 20% maximum voluntary contraction. For detailed information, please refer to our manuscript: https://doi.org/10.1101/2022.08.31.505855. The raw HD-EMG and corresponding CST are captured at 2048 Hz. We used a window length of 40 data points and sliding step of 20 data points with an overlap of 20 data points to segment the HD-EMG signals, and we used a window length of 20 data points and sliding step of 20 data points without overlapping to avoid double counting for the CST signal. --EMGs: n*40*64, where n is the number of frames, 40 is the window size of the signals, and 64 is the number of HD-EMG channels. These are the input signals to the deep CNN. --Spikes_cst: n*5, where n is the number of frames, and 5 is the maximum number of spikes allowed in one frame of HD-EMG signals. These are the output signals from the deep CNN. The Python implementation of the deep CNN can be found at https://github.com/ywen3/dcnn\_mu\_decomp.

本数据集包含高密度肌电图（high-density electromyography, HD-EMG）信号与对应的神经驱动信号，后者采用复合锋电位序列（composite spike train, CST）形式，用于训练和验证用于神经驱动估计的深度卷积神经网络（deep CNN）框架。本数据集纳入了5名受试者（uc4、uc8、uc11、uc25、uc26）的处理后数据，每名受试者对应两块肌肉——股外侧肌（vastus lateralis, VL）与股内侧肌（vastus medialis, VM），覆盖两次测试会话（d1与d2）。每次测试会话中，受试者完成3次等长收缩任务，收缩强度为最大自主收缩（maximum voluntary contraction, MVC）的20%。如需获取详细信息，请参阅本研究手稿：https://doi.org/10.1101/2022.08.31.505855。原始HD-EMG信号与对应的CST信号采样率均为2048 Hz。我们采用窗口长度为40个数据点、滑动步长为20个数据点（重叠20个数据点）的方式对HD-EMG信号进行分段；为避免重复计数，对CST信号则采用窗口长度为20个数据点、滑动步长为20个数据点且无重叠的分段方式。
-- EMGs：数据维度为n*40*64，其中n为帧数，40为信号窗口尺寸，64为HD-EMG通道数，该数据作为深度卷积神经网络的输入信号。
-- Spikes_cst：数据维度为n*5，其中n为帧数，5为单帧HD-EMG信号中允许的最大锋电位数量，该数据作为深度卷积神经网络的输出信号。该深度卷积神经网络的Python实现可在https://github.com/ywen3/dcnn_mu_decomp获取。