Siviero Ilaria, Helmhold Florian, Ray Andreas M, Bibian Carlos, Menegaz Gloria, Murguialday Ander Ramos, Storti Silvia F
Annu Int Conf IEEE Eng Med Biol Soc. 2024 Jul;2024:1-4. doi: 10.1109/EMBC53108.2024.10782268.
Technology for motor rehabilitation faces challenges in uncontrolled settings, such as at home. In these real-world scenarios, robust signals like electromyographic (EMG) and inertial measurement unit (IMU) data are crucial for decoding continuous human actions. Classical modeling methods, such as linear, adaptive, or static filters, lack the capacity to capture complex relationships between surface EMG and kinematics, as well as generalizability across subjects. We propose a deep learning (DL) model for the continuous decoding of hand motion. Custom-made wearable devices acquired EMG-IMU data from 27 healthy subjects performing a wrist flexion/extension task. Two regression models were compared with a hybrid convolutional neural network and a gated recurrent unit. To address inter-subject variability, a leave-one-subject-out cross-validation approach was implemented. The DL model showed a mean R increase of 0.18 compared to the polynomial regressor. Our approach enhances wrist kinematics decoding providing a generalized model based on data captured with wearable devices. The findings hold potential for innovative home-based telemedicine solutions in motor rehabilitation.
运动康复技术在非受控环境中面临挑战，例如在家中。在这些现实场景中，诸如肌电图（EMG）和惯性测量单元（IMU）数据等强大信号对于解码连续的人类动作至关重要。传统建模方法，如线性、自适应或静态滤波器，缺乏捕捉表面肌电图与运动学之间复杂关系的能力，以及跨受试者的通用性。我们提出了一种用于手部运动连续解码的深度学习（DL）模型。定制的可穿戴设备从27名健康受试者执行手腕屈伸任务时获取了肌电图 - 惯性测量单元数据。将两个回归模型与一个混合卷积神经网络和一个门控循环单元进行了比较。为了解决受试者间的变异性，实施了留一受试者交叉验证方法。与多项式回归器相比，深度学习模型的平均R值增加了0.18。我们的方法增强了手腕运动学解码，提供了一个基于可穿戴设备捕获的数据的通用模型。这些发现为运动康复中创新的家庭远程医疗解决方案具有潜力。
