Assessing the utility of Regenerative Peripheral Nerve Interfaces (RPNIs) in providing referred sensations in a functional task in a virtual environment

Abstract

Individuals who use upper limb prostheses receive limited feedback from their devices. Researchers have attempted to elicit sensation through direct stimulation of peripheral nerves or through stimulation of reinnervated skin or muscle. Previous research found that electrical stimulation of Regenerative Peripheral Nerve Interfaces (RPNIs) elicited sensations that were referred to the phantom hand. The purpose of this study was to determine if this sensation could be used to improve performance of a functional task. Two participants with upper limb loss completed the Box and Blocks Test in a virtual environment under four feedback conditions on a single day of testing. These conditions included no feedback, vibration triggered by object contact, and two conditions where RPNIs were electrically stimulated upon object contact. For the RPNI conditions, one was non-somatotopic, meaning the RPNI sensation was referred to an area on the phantom hand and the other was somatotopic, where the referred sensation and virtual sensor were collocated. Participants moved more blocks with somatotopic feedback to a moderate effect. Both participants expressed a preference for the somatotopic sensation, noting that it made their movements feel more natural. Overall, this study demonstrates that RPNI-elicited sensation has the potential to improve performance in functional assessments.

I. INTRODUCTION

To fully replicate the function of an intact hand, upper limb prostheses should allow for dexterous movement and naturalistic sensory feedback. Naturalistic sensations are often defined as those that have a similar quality (i.e., homologous) and occur in a similar location (i.e., somatotopic) to what is expected [1]. Currently, commercially-available prostheses offer only incidental feedback such as visual and auditory cues that the user interprets during the use of the device [2]. Incidental feedback alone is not sufficient to restore a prosthesis user’s prosthetic control to naturalistic levels, most notably in grasping tasks in which the weight of the object is unknown [3], [4].

To address these limitations, researchers have provided supplemental feedback to the user, most commonly through sensory substitution. In this approach, information from sensors on the prosthesis (i.e., contact pressure) is translated to the user through electrotactile, vibrotacticle, or mechanotactile feedback [5]. Common prosthetic sensor information includes joint position, grasp force, or pressure on an individual finger [4]. Sensory substitution has been shown to assist a user in identifying the location of an external force and the amount of pressure applied [6]. While beneficial, sensory substitution can also impose a significant cognitive burden as the user must learn to interpret the new stimuli as information associated with the state of their prosthesis [4], [5]. Additionally, this supplementary sensation only marginally improves functional task performance when visual feedback is also present [5].

In an effort to make sensory feedback more naturalistic (i.e., somatotopic), several research groups have instead focused on direct stimulation of peripheral nerves in the residual limb that once innervated the hand [7], [8]. These studies showed sensory feedback provided through the peripheral nervous system improves manual dexterity, prosthesis embodiment via increased stimulation selectivity, and reduces phantom limb pain [8], [9], [10]. Additionally, recording motor signals from the peripheral nerves improves prosthesis control [8], [9], [11]. Despite the improved selectivity of stimulation, recording efferent motor action potentials is difficult due to the small amplitude of the signals [12], and the longevity of these nerve interfaces is limited due to nerve inflammation or a foreign body response to neural electrodes [1], [13].

A surgical approach that prioritizes signal longevity and permits the generation of both efferent control signals and afferent sensation signals with high fidelity is the Regenerative Peripheral Nerve Interface (RPNI) [14]. An RPNI is constructed by wrapping a muscle graft around individually separated residual nerves or nerve fascicles. After allowing the muscle to revascularize and the nerve to reinnervate, RPNIs produce consistent motor signals with high signal to noise ratios [15] that can be used as input signals to provide accurate and stable control of a prosthetic hand [14], [16], [17]. Electrical stimulation of RPNIs produces sensations that are referred to the phantom hand [18], [19]. The location and quality of the evoked sensation is consistent over months to years [19].

Given that RPNIs produce efferent control signals and contain sensory afferents, it is possible that they can also be utilized in a closed-loop bidirectional feedback system. A prior case study found that an individual with transradial limb amputation using this system could classify four different objects by size and stiffness with 100% accuracy, when using vision and receiving feedback through stimulation of their RPNIs [19]. However, this work was limited to classification when an object was already in the hand. Thus, it is unclear how this sensation would affect performance of reach and grasp tasks.

Therefore, the purpose of this study was to determine the benefit of sensation elicited by RPNIs during a functional task. Unfortunately, few prostheses are equipped with contact sensors and where they do exist, they may not be located in the place that sensation is referred (i.e., sensors are in the fingertips but sensation is felt in the palm). To address this limitation, we instead assessed functional performance in a virtual reality environment (VRE). Our system enabled the user to control a virtual prosthesis using muscle activity while receiving sensation elicited by contact with virtual objects. This approach enabled us to directly compare the utility of sensation elicited by electrical stimulation of RPNIs to sensory substitution (i.e., vibrotactile feedback) and visual feedback alone. The elicited sensation was either somatotopic or non-somatotopic given where the stimulation of the RPNIs was felt by the participant. We expected that providing supplemental feedback in any form would improve functional performance over vision alone. We also hypothesized that somatotopic feedback would result in the highest performance.

II. Methods

A. Participants

Participants with unilateral limb loss at the transradial level provided their written informed consent to participate in this study, whose protocol was approved by the University of Michigan’s Institutional Review Board (HUM: HUM0012483; ClinicalTrials.gov #NCT03260400). Each participant had surgery to create Regenerative Peripheral Nerve Interfaces (RPNIs) on nerves in their residual limb. For some nerves, surgeons created multiple RPNIs from a single nerve after interfascicular dissection. Up to 12 bipolar electrodes (Synapse, Oberlin, OH, USA) were implanted into RPNIs and residual muscles either in the same surgery when the RPNIs were created or as part of a second surgery.

Participant 1 (P1) was a 51-year-old woman who underwent a distal transradial amputation due to infection. At the time of the transradial amputation, four RPNIs were constructed out of her median, ulnar, and dorsal radial sensory nerves. Her ulnar nerve was split to create two RPNIs (denoted Ulnar-1 and Ulnar-2). One year post-amputation, P1 had electrodes implanted in her one median and both ulnar nerve RPNIs, in addition to five residual forearm muscles. At the time of the experiment, P1 was 49 months post implantation.

Participant 2 (P2) was a 53-year-old man who underwent a wrist disarticulation of his left hand due to trauma. P2 later received targeted muscle reinnervation to address nerve pain, where his median nerve was split into two fascicles. He continued to experience pain and developed neuromas on his median nerve fascicles at the site of nerve coaptation. Two and a half years after the original amputation, P2 underwent surgery for the construction of four RPNIs from his median nerve. A surgeon placed electrodes in each RPNI and seven residual muscles as part of the same surgery. At the time of experimentation, P2 was 6.5 months post implantation.

B. Virtual Reality Environment (VRE) Overview

To assess the effect of feedback on manual dexterity, participants completed a series of trials of a VR representation of the Box and Blocks Test (BBT) under various sensory feedback conditions. The VRE facilitated the communication of several networked PCs, the Unity physics engine, and a bidirectional stimulation unit (Fig. 1). Upon a participant initiating an intended movement, efferent motor signals from surface electromyography (EMG) electrodes are decoded in real-time to determine the prosthetic control commands. These commands are sent to the PC running Unity, which renders the virtual environment and then actuates the virtual prosthesis within the environment. Contact events within Unity arise when the participant’s virtual hand touches the blocks or table and are then used to generate stimulation commands. These commands are passed from Unity to the stimulation unit, which produces stimulation waveforms that are sent to the appropriate RPNI via transcutaneous wire running to the electrodes implanted in the RPNIs.

Fig. 1.

Representation of how the participant receives RPNI-elicited sensory feedback and controls the prosthesis in virtual reality. During the trials, the participant used a motion tracking device to control the position of the virtual hand. Surface EMG signals were used to open and close the hand. Upon making contact with a virtual object, the force values were encoded into stimulation waveforms that were use to elicit sensation from the RPNI.

C. Prosthetic controller

For this experiment, the virtual prosthetic hand was controlled using surface EMG to decouple the stimulation from the controller. We first placed two sets of surface EMG gelled electrodes (EL503, Biopac, Goleta, CA, USA) on the participant, one set across the medial mass of muscle on the forearm and the other across the lateral mass of muscle approximately 4-6 cm distal from the lateral epicondyle of the elbow. Electrode placement was verified by visual inspection of signals when the participant contracted their residual muscles in a manner that corresponded with the perceived opening or closing of their phantom hand. EMG data were acquired and decoded using a neural signal processor (NeuroPort, Blackrock Microsystems, Salt Lake City, UT, USA) and real-time Matlab computer (xPC Target, Mathworks, Natick, MA, USA). EMG data were band-pass filtered to 100-500 Hz using a second-order Butterworth filter. We then calculated the mean absolute value (MAV) in 50 ms intervals. Grip aperture was then decoded in real-time using a position/velocity Kalman filter. The coefficients for the Kalman filter were calibrated by having the participants perform a bilateral mirrored training task, in which they attempted to match the position of their phantom limb to the position of a hand displayed on a computer screen. This method was based on previous RPNI control work, and a more in-depth explanation of the process can be found in [12]. Results from this training session were then processed offline and a decoder was built for online analysis. Correlation coefficients for the decoders were calculated, and if the linear correlation between the displayed hand positions and the Kalman filter predicted positions was less than 0.75, the electrodes were adjusted, and the participant completed another decoder training session. If the Kalman filter did not meet the 0.75 correlation threshold multiple times, two additional sets of surface electrodes were added. P1 required four electrodes, while P2 used two electrodes.

D. Virtual Box and Blocks Test (VBBT)

Participants completed a 20-minute training session to familiarize themselves with the VRE. During this training, participants donned the VR headset and completed a virtual version of a standard clinical assessment, the BBT [20]. In the physical BBT, participants are placed in front of a box divided by a wooden partition. One side of the partition holds 150 2.5 cm³ wooden blocks placed in random orientations. The participant’s score is equal to the number of blocks moved over the divider and dropped to the other side. The Virtual Box and Blocks Test (VBBT) similarly consists of two regions separated by a partition, with the dimensions of the test being identical to the BBT. Two critical changes were made to differentiate the VBBT from the physical version. First, we lowered the walls containing the blocks so that the virtual hand was less likely to get caught on a wall, and second, we reduced the number of blocks to 18 to reduce the computational load of the simulation.

Participants completed the VBBT under various feedback conditions, in random order (Table. I). In one condition they had visual feedback from the VR, but no supplemental feedback (‘No Feedback’ condition). In the ‘Vibrotactile’ condition, sensors in the virtual hand would trigger the vibration of a second controller strapped to the participant’s upper arm. The next two forms of feedback involved electrical stimulation of RPNIs using a portable stimulation unit (NOMAD, Ripple Neuro, Salt Lake City, UT, USA). The maximum stimulation amplitude of this system was 2.5 mA. In a previous study, we used a different system to obtain sensory thresholds with a stimulation range of 10 mA using an adaptive staircase approach [19]. Based on these results, we selected RPNIs with the lowest amplitude sensory thresholds for providing feedback. Electrical stimulation was provided such that the feedback was somatotopic (location-matched) or non-somatotopic. The Somatotopic condition involved the placement of virtual sensors only in locations where the participant had previously experienced referred sensation during RPNI stimulation (Fig. 3). These sensors then triggered stimulation of the corresponding RPNI. Non-somatotopic sensation only utilized virtual sensors placed at the distal end of the index finger and thumb of the virtual prosthesis (the main points of contact to pick up a block). These pre-specified sensors then triggered stimulation of RPNIs selected based on their proximity and position relative to the sensor positions on the hand. For P1, the sensor on the index finger would trigger stimulation of her Ulnar-1 RPNI and the sensor on the thumb would trigger stimulation of her Median RPNI. Stimulation was always delivered via symmetric, square, charge-balanced, biphasic waveforms with timeinvariant stimulation parameters. Stimulation waveforms had a 2.5 mA amplitude, 100 μs pulse width and 20 Hz pulse frequency.

Table I:

Conditions for the Virtual Box and Blocks Task

Condition	Sensor Location	Stimulation Method
No Feedback	N/A	N/A
Vibrotactile	VR Fingertips	Vibration of the Upper Arm
Non-Somatotopic	VR Fingertips	Electrical Stimulation of the RPNI
Somatotopic	Perceived Sensation Area	Electrical Stimulation of the RPNI

Fig. 3.

Virtual sensors were created to match the location the individual felt sensation when their Median and Ulnar RPNIs were electrically stimulated. This represents the Somatotopic condition.

P1 completed the task in the order: Non-Somatotopic, Somatotopic, No Feedback, and Vibrotactile trial. P2 completed the task following the order: No Feedback, Vibrotactile, and Somatotopic trial. Only P1 completed the task under Non-Somatotopic condition since the stimulation of P2’s Median-4 RPNI was already location matched with the sensors on the virtual hand. At the completion of the 20-minute training session and after the participants had become familiarized with each feedback condition, they were given a 5-minute break before experimental trials started. For each trial, participants had one minute to move as many blocks as possible from one side of the partition to the other. Participants completed three trials per feedback condition, with a 1-minute break between each trial and a 10-minute break with the headset off between conditions.

During testing, participants expressed that sensation changed the way they used their muscles to complete the task. To assess this, we quantified total activation of the wrist extensor and flexor muscles over each one minute VBBT trial. Real-time EMG data were collected at 30 KHz, band-pass filtered to 100-500 Hz, rectified, and then the MAV at each millisecond was determined over a 50 ms window. This effectively downsampled the data to 1000 Hz before saving it to the computer. We then calculated the EMG linear envelopes using a 40 ms RMS moving average filter (Fig. 2). Finally, we calculated the integral of the EMG data over each one minute trial as a measure of the total muscle activation.

Fig. 2.

EMG linear envelopes calculated from Participant 2’s first trials of the No Feedback, Vibrotactile, and Somatotopic conditions. The linear envelopes were calculated using a 40 ms RMS moving average filter across the minute trial. The integral of the EMG linear envelope was used as a measure of the participant’s total muscle activation during that trial. Both participants expressed that sensation changed the way they used their muscles during the VBBT. Thus, total muscle activation was used in combination with the number of blocks moved during the trial to quantify this change.

E. Statistical Methods

The primary dependent measure was the number of blocks moved in each trial. As a secondary outcome, we compared the muscle activity between conditions. Given the potential differences in the number of blocks moved, we calculated the total activity per block moved as the ratio of the integrated EMG over the full minute trial to the number of blocks moved during that trial. To combine different muscles, the integrated EMG data for each muscle group across all conditions were standardized to a mean of 0 and standard deviation of 1. Here we used an n of 1 design to compare conditions within an individual. For each participant, we compared the number of blocks moved (n = 3 trials per condition) and muscle activity per block transferred (n = 6-12 integrated EMG recordings for 3 trials per 2-4 muscles) between each feedback condition and the No Feedback condition using Hedge’s g effect sizes. Given the small sample size and exploratory nature of the study, we highlight effects that are small (g ≥ 0.2) medium (g ≥ 0.5) or large (g ≥ 0.8) [21].

III. Results

A. Sensation location

P1 had two RPNIs, the Median and the Ulnar-1, consistently refer sensation to the phantom limb at stimulation amplitudes below the 2.5 mA threshold. P1’s Median RPNI referred sensation to her thenar eminence, while her Ulnar-1 RPNI referred sensation to the side of her hand proximal to her small finger (Fig. 3). P2 had a single RPNI, Median-4, consistently refer sensation to the phantom limb below the 2.5 mA threshold. P2’s Median-4 RPNI referred sensation to the tip of his index finger, either on the pad of the fingertip or along its edge. Sensation for both participants was described as unnatural, but not uncomfortable tingling.

B. Virtual Box and Blocks (VBBT) Performance

P1 completed the VBBT under all feedback conditions, moving an average of 4.5 blocks. (Fig. 4). There was a medium effect for the difference in blocks moved during the No Feedback and Somatotopic conditions (g = 0.651) and a small effect for the Non-Somatotopic condition (g = 0.326). Compared to the No Feedback trial, P1 moved an average of 0.73 more blocks during the Somatotopic condition (15.38% improvement) and 0.33 blocks fewer during the Non-Somatotopic condition (7.62% reduction). P2 did not complete the Non-Somatotopic condition, as the sensors for the Somatotopic condition overlapped with the locations of the sensors for the Non-Somatotopic feedback condition. There were moderate effect sizes for the differences in all remaining feedback conditions. P2 moved an average of 1.77 more blocks (17.89% improvement) during the Vibrotactile trials (g = 0.694) and an average of 3.33 more blocks (35.79% improvement) during the Somatotopic trials (g = 0.790).

Fig. 4.

Number of blocks moved during the Virtual Box and Blocks Test. Participant 1 completed the task under all four feedback conditions. Participant 2 did not have the Non-Somatotopic condition as his sensation was referred to the fingertip. Conditions where the comparison to No Feedback had a moderate effect size are denoted by ††, and conditions with small effect sizes are denoted by †.

The participants would occasionally fail to transfer a block. Missed block transfers by either participant were primarily due to lifting the hand up without successfully grabbing the block or opening the hand before transferring the block over the divider. P1 made the former error more frequently, due in part to her difficulty in opening the virtual prosthesis during some periods of the collection. P2 more commonly made the latter error because he was moving quickly and would lose a hold on the block.

C. Participant Feedback

Both participants were asked to describe their subjective experience engaging with the VRE and using a virtual bidirectional prosthesis. P1 primarily described sensation as making the VBBT easier, even if it did not necessarily improve her performance. Of the task, she said “It was easier [with sensation] because it let me know where the blocks were. Like, oh, okay, I really am touching a block” and “It lets me know I have something. Like I can feel the object”. P2 typically describes his phantom limb as “locked solid”. He reported that “For me to open it, I’m really stressing… but while I was doing [the task] it felt pretty normal”. During one trial using somatotopic feedback, P2 explained “My hand is in cement all the time, but when I am doing this… I feel like I am actually opening and closing it”. He went on to say, “It wasn’t that the feedback was making me better because I had visual, but the feedback was making it comfortable. It made me do better” and “I felt my finger on it, I felt my hand, literally doing this” as he made opening and closing gestures with his intact hand.

D. Muscle Activation per Blocks Moved

P1 had greater muscle activity per successful block transfer during the Non-Somatotopic condition compared to receiving no feedback, with a large effect size (g = 1.851; Fig. 5). In contrast, P1 had lower muscle activity during the Somatotopic condition (g = 1.071). There was no difference between the No Feedback and Vibrotactile trials (g = 0.090). P2 had lower muscle activity during both the Vibrotactile condition (g = 0.905) and Somatotopic condition (g = 0.973) compared to No Feedback.

Fig. 5.

Muscle Activation per Block Successfully Transferred: integrated electromyography signal divided by the number of blocks moved during the trial for each condition. Participant 1 had four surface electrodes, two targeting the wrist flexor muscles, denoted as Flexors Group 1 (Flexors G1) and Flexors Group 2 (Flexors G2) and two targeting the wrist extensor muscles, denoted as Extensors Group 1 (Extensors G1) and Extensors Group 2 (Extensors G2). Participant 2 had one surface electrode on the wrist flexor muscle group and one on the wrist extensor muscle group.

IV. DISCUSSION

The purpose of this study was to determine how RPNI feedback affected performance during a functional task. Both vibrotactile and somatotopic feedback did improve performance with medium effect sizes for P2. For P1, the vibrotactile feedback had a small effect and the somatotopic feedback had a medium effect on performance. Future studies, including a larger number of trials or increased training time, may therefore demonstrate statistically significant differences.

Given the novelty of the task, it is difficult to determine what level of change in performance would be clinically meaningful. Minimal detectable change values have been reported for the physical BBT [22], but not for studies which included individuals with upper limb loss. Additionally, studies of other virtual reality BBTs, have found that while their virtual BBT outcomes were correlated with the physical BBT, the relationship was not 1 to 1 [23], [24]. Participants tend to move 1.5-2.5 times more blocks on average in the physical BBT compared to virtual one [23], [25]. These studies also used different approaches to grasping blocks (e.g. gesture control [23] or a trigger on a motion controller [25]) and studied different populations (e.g. stroke [25] and Parkinson’s [23]). Prior studies also assumed the block was lifted when the hand was near or the trigger was pulled and did not incorporate physics models. We expected that using muscle activity to control the virtual hand and using a physics model for hand/block interaction would be more difficult than these approaches and increase the likelihood of dropped blocks. Indeed, we measured the fewest number of blocks moved across studies.

Others have explored the effect of feedback on the performance of the physical BBT with a prosthesis [26], [27]. In one study, six subjects received vibrotactile stimulation while using a myoelectric hand prosthesis to complete a series of functional assessments [26]. The subjects’ performance during the No Feedback and Feedback conditions of the BBT were not significantly different. However, it was found that vibrotactile feedback only benefited the user during more complex assessments such as the Clothespin Relocation Test (CRT) and the Cups Relocation Test (CUP) which require the subject to dexterously manipulate and move an object.

Interestingly, this study [26] also found that some users only demonstrated functional improvement during the Feedback trials after they learned to accurately control the myoelectric prosthesis. In the CUP task, the participants learned to more reliably grab the cups at lower force levels with their myoelectric prosthesis. This translated to a significant performance improvement between the two sessions when vibrotactile feedback was provided. For our study, this suggests that the novelty of learning to control the VR hand with surface EMG may have affected the efficacy of the Non-Somatotopic and Somatotopic feedback conditions. In fact, there is some evidence that our participants, especially P1, may have experienced some difficulties with controlling the virtual hand. While P1 had a higher correlation coefficient (0.93) compared to P2 (0.78), P1 required two sets of electrodes and nearly two hours of additional time to train a viable decoder. P2 only required a single attempt to complete the decoder build and ultimately completed nearly twice as many block transfers compared to P1 in each trial. Additionally, P2 used a myoelectric prosthesis at home, whereas P1 used a body-powered prosthesis at home and had only controlled myoelectric prostheses in an experimental setting.

Between each trial, each participant was asked to convey their experience using the current feedback type. P1 did not prefer a specific type of sensation, but expressed that feedback made the functional assessment easier because she knew when the VR hand was contacting the block. On the other hand, P2 noted that he was the most comfortable performing the VBBT during the Somatotopic condition. He explained that with somatotopic feedback, he could feel his phantom hand opening and closing, whereas it normally feels stuck in cement and is unresponsive. He felt that his performance improved due to this new sensation. Positive emotional responses to feedback have also been reported in other studies of peripheral nerve sensations [28], [29]. However, our participants’ responses may be biased by the fact that the sensation was new to them and they had limited exposure to it. Future studies will need to explore whether this form of sensation is viewed positively when used more regularly in an individual’s life.

We interpreted P2’s comments to indicate that he was able to relax muscles that he felt were usually continually activated. To explore this idea, we compared the total muscle activity across conditions. Given the differences in the number of blocks moved per trial, we divided the recorded activity by the number of blocks successfully moved. In this sense, we measured the muscular “effort” used to move a block. Both participant had decreased muscle activity per moved block during the Somatotopic condition compared to No Feedback, with moderate to large effect sizes (Fig. 5). P2 also responded to Vibrotactile feedback by decreasing activity for each block moved, while P1 did not. P1 also completed the Non-Somatotopic condition, which actually resulted in the highest muscular activity per block moved. In this condition, sensation of the fingers contacting the block was referred to the locations of her referred sensation, which was in the palm. This effectively makes the Non-Somatotopic condition a different form of sensory substitution, which is potentially even more challenging to interpret due to its presence in her phantom hand and not on her residual limb. P1’s lack of benefit during the Vibrotactile condition further suggests difficulty incorporating sensory substitution approaches. It is possible that performance might improve over time as sensory substitution can be learned [26].

This study has several limitations. First, the study only involved two participants, which limits the generalizability of our findings. Our investigation was also limited by the specific areas of sensation that our participants could experience. Currently, RPNI-elicited sensation is limited to a single potential perceived area per RPNI, and our bidirectional system limited the potential sensation areas further by restricting the experiment to only use RPNIs that could consistently evoke referred sensation at 2.5 mA or lower. This resulted in two usable RPNIs for P1, and a single usable RPNI for P2. With a system that either allows for stimulation at higher amplitudes, or participants with RPNIs that evoke sensation at different areas within the amplitude limit, we may see different responses when provided with referred sensation. This study was conducted in a single session. While that provides consistency in the controller utilized, it limits our ability to assess training effects. It is possible that the feedback provided takes time to incorporate and differences between conditions would be more pronounced over time.

V. CONCLUSIONS

In this study, we demonstrated the effect that sensation elicited from RPNI stimulation has on the function of an individual using a virtual prosthetic hand to complete the VBBT. Both participants moved more blocks and used less muscle activation to move each block during the Somatotopic trials. The participants found the introduction of somatotopic sensation to be enjoyable and useful. Their testimonials also reveal how referred sensation improves their ability to control the virtual prosthesis. One participant experienced a change in how their phantom hand moved, with their phantom hand opening and closing more responsively once they started completing the VBBT with somatotopic sensation. Ultimately, this pilot study provides insight into the value RPNI-elicited sensation has on a person’s control and enjoyment of a prosthesis. Future studies will continue to assess the effect of somatotopic sensation on a larger population and will incorporate RPNI-elicited control, as well as sensation to the prosthesis user.