High-Density Electromyography-Based Methods for Joint Torque Prediction and Motor Unit Behavior Observation
Time: Tue 2023-03-07 10.00
Location: E32, Lindstedtsvägen 3, Stockholm
Subject area: Engineering Mechanics
Doctoral student: Asta Kizyte , Teknisk mekanik
Opponent: Assoc.Prof. Christian Antfolk, Lunds universitet
Supervisor: Ruoli Wang, BioMEx; Elena Gutierrez-Farewik, Teknisk mekanik, BioMEx
Electromyography (EMG) is a technique that measures the electrical activity of muscles. It reflects muscle activation and provides an interface to the central nervous system at the level of the muscle or individual motor units, which helps us understand the mechanisms of muscle force production, control, and coordination. EMG can also be used to detect changes in muscle activity caused by pathology, making it a valuable tool for research, diagnosis, and rehabilitation. One of the latest advancements in EMG technology is high-density EMG (HD-EMG). HD-EMG measures multiple spatially separated samples of muscle activation. This additional spatial information in HD-EMG offers new possibilities for the prediction of joint torques and the ability to look into individual motor units by decomposing the signals using blind source separation methods. This thesis presents two studies that explore the use of HD-EMG methods for joint torque estimation and the observation of motor unit behavior.
In the first paper, we presented a detailed investigation of the effects of different EMG and kinematic inputs on the accuracy and robustness of ankle joint torque prediction using support vector regression. To evaluate the robustness, we analyzed the results in three cases (intra-session, inter-subject, and inter-session) and two movement categories (isometric contraction and dynamic movement). We found that HD-EMG-derived inputs improve the accuracy and robustness of torque prediction of the isometric contractions. However, in dynamic movements, good prediction results could only be achieved by including additional kinematic features (ankle joint position and angular velocity), and the type of EMG input did not strongly influence the results.
In the second paper, we investigated the changes in motor unit behavior of the ankle plantar flexor (soleus) and dorsiflexor (tibialis anterior) caused by spinal cord injury (SCI). We computed torque, EMG, and motor unit parameters during volitional sub-maximal voluntary contractions for the SCI group and compared them to a non-injured control cohort. We found that participants in both groups could maintain the prescribed torque with similar variability. However, the SCI group required higher muscle activation levels (normalized to maximum) to achieve the same level of relative torque compared to the control group. The SCI group had lower intramuscular coherence in the alpha frequency band than the control group, indicating altered neural synchronization at the sub-cortical level. The soleus motor unit firing patterns were more variable post-SCI than in the control group. In addition, at high torque levels (50% of personal maximum), both muscle's motor units were recruited and de-recruited at lower torques, and motor units fired at lower rates in the tibialis anterior muscle in persons with SCI, indicating altered force gradation strategies after the injury.
The studies presented in this thesis demonstrated that HD-EMG is suitable for robust isometric ankle joint torque prediction, which has potential in applications such as robot-assisted rehabilitation and robotic gait assistive technology. In particular, the robustness and accuracy of HD-EMG-based predictions are essential for improved estimation of the joint torque that can then be used in the human-in-the-loop control scheme. In addition, HD-EMG decomposition enables a non-invasive way to observe the motor unit behavior in vivo in persons with neuromusculoskeletal disorders, which can enhance the understanding of the underlying neurophysiological mechanisms of motor impairments. The insights provided by such HD-EMG analysis in the future may be beneficial for developing targeted interventions and personalized therapies.