(I-321) HOW TO IMPLEMENT WEARABLE ULTRASOUND FOR PROSTHETIC HAND CONTROL

Introduction:: Advanced upper limb prostheses, such as multi-degree-of-freedom prosthetic hands, are controlled by inferring a user’s intentions (e.g., to grasp an object) to derive the control signals sent to the prosthesis actuators. Inferring user intent is typically done by assessing the residual muscle activities which arise from the descending motor commands that control the limb. Electromyography has become the primary clinical approach for evaluating residual muscle activity, which uses surface electrodes placed over the residual musculature [1]. However, surface electrodes can only examine a portion of the descending motor commands that control the residual limb, because the electrical activities of deep musculature are either not detected or diffused as cross-talk. Sonomyography, the ultrasound-based sensing of muscle deformation, is an emerging alternative sensing modality for upper limb prosthesis control [2]. Sonomyography enables spatiotemporal characterization of both superficial and deep muscle activity, making it possible to distinguish the contributions of a larger set of muscles when deriving prosthesis control signals. Previously, sonomyography was not feasible for prosthesis control due to high power requirements and large transducer sizes. However, our research lab has recently incorporated a new ultrasound imaging approach that borrows from techniques in the radar literature, which enables miniaturization of ultrasound instrumentation using low-voltage commodity hardware and low-frequency processing speeds [3]. We present a 4-channel wearable ultrasound system capable of tracking in vivo muscle interfaces that can feasibly be used to control a prosthetic hand.

Materials and Methods::
Our implementation employs frequency-modulated continuous wave (FMCW) ultrasound imaging instead of conventional pulse-echo approaches. A key feature of FMCW imaging is the use of a linear chirp signal to encode the depth of ultrasound reflections as a range of frequencies, which bypasses the need to transmit short-duration high amplitude pulses to create a depth-resolved map of muscle tissue. Our system consists of a chirp generator, four single element ultrasound transducers (7mm diameter, 0.5 mm thick), a power regulation subsystem providing ± 5 V, and microprocessor for signal processing. Our system uses M-mode ultrasound images to assess the motion and timing of the internal structures, which are used to derive the prosthesis control signals. We are currently using machine learning to recognize the patterns of user intent (analogous to pattern recognition for myoelectric control).




Results, Conclusions, and Discussions:: We anticipate that our implementation of low-power ultrasound imaging will serve as the foundation for future prosthetic and exoskeleton designs. One of the primary benefits of the sonomyographic sensing modality is that muscle activity can be sensed with high spatial specificity, even in deep-seated muscle compartments. This presents an opportunity to derive more independent control signals than electromyographic sensing can provide. Our previous benchtop testing with able-bodied participants revealed that sonomyography can distinguish up to 15 unique hand grasps with 91% cross-validation accuracy [3]. The high spatial and temporal resolution of our approach shows promise for generating reliable control signals that are intuitively executed by the user and require minimal training. We found that transradial amputees could achieve 96% classification accuracy for 5 unique hand grasps after only a few minutes of training the algorithm [3].

Wearable ultrasound can now feasibly be implemented to control a prosthesis by analyzing patterns of residual muscle activities. However, this technology could be applied to a broad range of applications, including exoskeleton control, muscle monitoring, wearables, and biofeedback devices.



Acknowledgements (Optional): :
This study is supported by NIH award 5U01EB027601 and CDMRP RESTORE award W81XWH2010817.




References (Optional): : [1] Resnik et al. (2018), Neuroeng. & Rehab.; [2] Engdahl et al. (2022), Front. Bioeng. & Biotech.; [3] Acuña et al. (2022), Myoelec. Contr. & Upper Limb Prosth. Symp.