Two transradial amputees (amputee 1: a 55-year-old male with electric shock amputation in 1989, amputee 2: a 60-year-old male with explosion amputation in 1980) were recruited for this research. Five able-bodied subjects (1 male, 4 females, 20∼25 years old) were recruited. All subjects met the following requirements: (a) not taking drugs that affect hormones or neurotransmitters in the last 30 days, (b) no electromagnetic hypersensitivity, (c) no psychiatric or cognitive disorder, and (d) experience using a myoelectric prosthesis. The experimental procedure was approved by the Chongqing University Three Gorges Hospital Ethics Committee (2021-KY-24). All subjects signed informed consent forms before the experiments, which includes the stimulation and prompts they would receive and what operations they needed to perform in the experiment.

The experimental platform mainly includes a PC, a control module, an upper-limb prosthesis and other devices, as shown in Figure 1. We independently designed, drew, and welded the control module and integrated the parts above to perform the following experiment.

The parameters of the biphasic current waveform are adjustable (orange dashed rectangle in Figure 1): frequency (reciprocal of period) = 100∼500 Hz (100 Hz increments), pulse width = 100∼500 μs (100 μs increments), delay = 100∼500 μs (100 μs increments), current amplitude = 0∼8 mA (0.25 mA increments, 5 mA max for position and movement sense experiment), and burst duration = 0.5∼1 second (100 ms increments).

All subjects were required to sit on a chair in a comfortable posture; the able-bodied subjects’ dominant arms were placed on a sponge pad, and the plane of the palms was perpendicular to the ground. Amputees placed the residual limb on a sponge pad as well and were asked to keep the phantom palm in a straight (ST) position. For consistency, the circumference of 10–12 cm above the styloid process of the ulna and 2–4 cm above the amputation end were the places where able-bodied and amputees attached stimulation electrodes, respectively. A reference electrode was attached to the olecranon for each subject. CH1 is on the volar side, and eight channels were equally attached and arranged along the pronation direction. The connecting line of the centers of eight circular electrodes formed a plane perpendicular to the connecting line of the wrist and elbow (Figure 1).

First, we conducted a preliminary experiment to explore the forearm skin sensory threshold and discomfort threshold of each subject. Referring to the general experimental paradigm of electrotactile evoked sensation (Chai et al., 2015; Zheng and Hu, 2018), we fixed the frequency of all electrical stimulation at 200 Hz, the pulse width at 500 μs, and the delay at 100 μs. Taking the ith channel as an example, the current amplitude was incremented from 0 μA in steps of 500 μA. Each stimulation lasted for 1 s and was then followed by a 10 s rest period. Once the stimulation was perceived, it was repeated 3 times to ensure that subjects perceived the stimulation clearly. The stimulation amplitude was recorded as the sensory threshold A_i1–up if perceived and then increased until the subjects felt discomfort, and the current amplitude was recorded as the discomfort threshold A_i2. The current value was set to 40% of the maximal current, then it decreased in steps of 250 μA until the subjects could not perceive the stimulation. The last current value was recorded as A_i1–down. The maximum value between A_i1–up and A_i1–down was considered the sensory threshold A_i1. In addition, subjects were asked to describe the perceived sensations of stimulation, such as pressure, vibration, numbness, and pain, during this experiment. The current amplitude of each channel was fine-tuned by comparing the sensory threshold in neighboring channels to achieve similar tactile sensation across channels (Garenfeld et al., 2020).

The study chose spatial coding to feedback the positions of wrist flexion and extension (FE), because spatial coding is easier for recognition than intensity coding or temporal coding. As shown in Figure 2, we primarily chose five angular positions of wrist FE with 30° of resolution, which were named extension 60° (E60), extension 30° (E30), ST, flexion 30° (F30), and flexion 60° (F60). As shown as attachment of electrodes in Figure 1, active electrodes among eight electrodes around the forearm correspond to the five positions of wrist FE. The position of active channel corresponds to the direction of wrist FE movement. As the limit positions of wrist FE movement, E60 and F60 are configured with single channel of active electrode. For example, the CH5 on the dorsal forearm corresponds to the limit extension direction of the prosthetic wrist, so it is coded as E60 position. The CH1 on the volar forearm corresponds to the limit flexion direction of the prosthetic wrist, so it is coded as F60 position. As the initial position of wrist FE movement, ST is configured with dual channels of active electrodes. The dual channels are related to CH7 and CH3. As the non-limited positions of wrist FE movement, E30 and F30 are configured with dual channels of active electrodes. For example, the CH 6 and CH4 close to the dorsal forearm corresponds to the extension position of the prosthetic wrist, so it is coded as E30 position. The CH 8 and CH2 close to the volar forearm corresponds to the flexion of the prosthetic wrist, so it is coded as F30 position. The recognition of a single channel is easier than that of dual channel (Geng et al., 2016), so we related the single channel to the limit position of wrist FE, informing the subject that the prosthetic wrist has moved to the limit position. The burst duration for each stimulation mode is fixed at 0.5 s. Before the electrical stimulation was executed, the prosthetic wrist moved to the corresponding position.

This experiment was composed of two sessions: a learning session and a test session. A learning session was arranged before the test session to familiarize the subjects with the electrotactile scheme. The stimulation modes occurred randomly. At the same time, the stimulating channel map and the corresponding prosthetic wrist state were displayed on the screen (Figure 3A). After 0.5 s of stimulation, there was a 10-s rest period. The learning session lasted approximately 10 min. After subjects passed an evaluation of learning, the test session would be executed.

The paradigm of the experiment in the test session is shown in Figure 3B. Each session contains 30 blocks (5 kinds of wrist FE positions *6 blocks), and each wrist FE position accounts for 6 blocks. Each block consisted of three identical trials. Each trial consisted of 2 s of preparation time during which a beep sound was played to prompt the subjects, a 0.5 s stimulation period, time for the participant to answer, and 5 s of rest time. Then, 30 s of rest was used to relieve mental fatigue. There was no reminder (including audition) in the test session. When waiting for the answer of subjects, a dialog box popped up on the screen. The subject needed to give the index of the prosthetic wrist FE position corresponding to the stimulation in the dialog box as fast as possible by pressing a single number and the enter key on the keyboard. The DRT was counted from the end of the stimulation to the time when the enter key was pressed. If there was no answer within 20 s, the trial was considered to be a failed recognition.

Based on the static position sense experiment, the study selected initial and end position from five wrist positions to form a movement mode. Our study chose eight movement modes from 20 combinations (5*4). Eight movement modes are: extension small 1 (ES1), extension small 2 (ES2), extension small 3 (ES3), extension large (EL), flexion small 1 (FS1), flexion small 2 (FS2), flexion small 3 (FS3), and flexion large (FL) (Figure 4). The variables include direction (F or E) and range (S:60° or L:120°). When the prosthetic wrist moved within the perception range of the five positions, the corresponding stimulation channels were active. After the prosthetic wrist moved in the next range of the preestablished positions, the previous channels were inactive. For consistency with the static position sense experiment, the burst duration on each electrode also lasted for 0.5 s. Therefore, the total duration of stimulation of ES1, 2, and 3 and FS1, 2, and 3 was 1.5 s, while that of EL and FL was 2.5 s.

The movement experiment also includes learning and test sessions. During the learning session, the subjects were provided with three kinds of guidance: 1. the movement of the prosthetic wrist, 2. the stimulating channel map, and 3. the dynamic illustrations of wrist FE movements on a 22″ screen. The above guidance helped subjects establish the connection between electrotactile stimulation and wrist movement to achieve a better learning effect. The learning session lasted approximately 15 min. The paradigm of the test session is similar to that of the static position sense experiment. The test session consisted of 32 blocks of random movement modes. To simplify user input, we numbered E60∼F60 as indexes 1∼5. Similarly, the subjects were asked to respond to the perceived movement by a keyboard. They needed to press the first index on the keyboard to represent the initial position, then the cursor was automatically switched to the next dialog box in which the second index represents the end position, and they finally pressed the enter key to submit the answer (Figure 3C). Similarly, we recorded the input and DRTs of the subjects. For consistency of the stimulation electrode position on the forearm in the two experiments, we used multiple reference positions such as the styloid process of ulna, olecranon, etc. In addition, we took photos of each subject’s forearm and marked the position of each electrodes with a color pen that is harmless to the skin.

Subjects’ subjective feelings need to be considered. To evaluate the acceptance of the electrotactile scheme in this study, after each subject completed the movement experiment, we distributed a questionnaire and invited the subjects to use a score from 1 to 5, where 5 represents the highest outcome, to rate the following aspects: 1. degree of pain and numbness, 2. the resolution of each channel, 3. the comfort of electrotactile stimulation, 4. intuitiveness, and 5. ease of learning.

We analyzed the sensory and discomfort thresholds and SR of each stimulation mode in two experiments. Specifically, in the movement experiment, we analyzed the SR from the following two aspects. (a) FE direction and range. Eight movements were divided into four categories (ES, FS, EL, and FL) by FE direction and range. For example, the given stimulation is ES2, while the answer is ES1, ES2, or ES3. This counts as a successful recognition in this aspect. (b) Each mode: only when both the initial position and end position were correctly identified can it be counted as a successful recognition. The DRT of success and wrong recognition of each stimulation were analyzed. The non-parametric Kolmogorov-Smirnov test (K-S test) and Bonferroni test were performed to detect the difference in each channel of sensory threshold and discomfort threshold and DRT, since previous tests have shown that all statistics failed to pass homogeneity of variance. The threshold for statistical significance was set at p < 0.05. Statistical analysis and graphing were performed in Prism 8.0.2 (GraphPad Software Inc, CA, USA).

In the preliminary experiment, the sensitivity of eight electrode channels was measured in all subjects. Differences in sensory thresholds between volar forearms and dorsal forearms of able-bodied subjects (Figure 5) may be due to anatomical structures. The average sensory thresholds on the volar side (such as for CH1, CH2, and CH8) of able-bodied subjects were lower than those on the dorsal side (CH4 and CH5). Transcutaneous electrical stimulation can not only act on skin to induce superficial sensation but also activate afferent sensory nerves. If the nerve distribution in this area is more superficial, the threshold is lower. The muscle spindle in the middle of the muscle is in a relatively superficial position under the skin, which results in the lowest mean sensory threshold of its sensory nerve. The muscles at the CH4 position are relatively thick, and the muscle spindles in the middle of the muscles are deeply distributed under the skin, resulting in a higher sensory threshold. Stimulation close to the median nerves may induce uncomfortable numbness. Therefore, we tried to avoid this volar forearm area or reduce the amplitude in these channels.

The difference in electrotactile sensitivity between the amputees and the able-bodied subjects was the sensory threshold (Figures 5A, C). The sensory thresholds in the two amputees were higher than those of the able-bodied subjects. The higher sensory threshold indicates that the nervous system of amputees needs to be injected with more stimulation energy to produce a similar sensory type to that of healthy subjects. The results reflected that amputee’s sensitivity is decreased than healthy subjects. This may be due to the sensory nerve impairment caused by transradial amputation. Kosasih and Silver-Thorn (1998) showed that unilateral tibial amputation caused superficial pain, vibration, and/or impaired touch sensation. In addition, electrotactile sensation is also relevant to mechanoreceptors distribution (He et al., 2016), reason and time of amputation, the age of subjects, etc. The discomfort threshold of CH1 was rather low in amputee 1. This might have been caused by the amputation operation performed on this subject. The skin sensitivity of the residual limb varies widely in space. Therefore, it is necessary to set the amplitude current independently on each channel.

The real current amplitude in each subject was fine-tuned around the sensory threshold to induce similar tactile sensation. Considering that long-term stimulation can induce sensory adaptation and numbness, we chose a relatively small amplitude for each channel, but this may have decreased the channel discrimination.

The study employed five kinds of stimulation modes to code the FE position of the prosthetic wrist, and the study also employed eight kinds of stimulation modes to code the FE movement of the prosthetic wrist.

For position sense, the test results in all able-bodied subjects preliminarily demonstrated the effectiveness of the electrotactile scheme. Upon further analysis of the wrongly recognized positions, dual-channel stimulation mode (F30, ST, E30) often reduced the difference between two channels, such as in the F30, F60, and ST positions. In our limited results, the SR of amputees was not inferior to that of able-bodied subjects. The possible reason is that amputation leads to different neural plasticity outcomes between amputees and able-bodied subjects with intact proprioceptive circuits (Di Pino et al., 2009; Terlaak et al., 2015). The overall SR of amputee 1 was better than that of subject 2. The slightly higher learning ability of subject 1 is a possible reason for this result, and the influencing factors might include subject 1’s younger age, higher education level, and higher economic level. Subject 2 did not clearly distinguish between ST and F30 positions. The possible reason for this result is that since the upper arm is not an ideal cylinder, CH3 and CH7 are close to CH2 and CH8. The induced sensations may also be quite similar, making them difficult to distinguish. This wrong recognition caused by spatial adjacency was also found for able-bodied subjects.

For movement sense, a temporal combination of multiple position sensations, hence the subject needs to perceive not only spatial change but also temporal change (F direction or E direction). Therefore, we predicted that the success rate of movement sense is lower than that of position sense. The results are consistent with the prediction. All subjects had a high SR for both the direction and the range of movement. Amputee 2 acquired a better SR in large-range FE movements than in small-range movements. For the same speed of movement in the prosthetic wrist, the burst duration of a large-range movement is longer than that of a small-range movement. Therefore, subjects can easily distinguish the large-angle range and the small-angle range at the time of stimulation. From the results of each stimulation mode, ES2, FS2, and FS3 were poorly recognized by able-bodied subjects. We speculate that the reason for this outcome is that the three modes all included position-F30, which was poorly recognized by able-bodied subjects in the position sense experiment (58.89%). The SR of amputee 1 was not lower than that of the able-bodied subjects, which may be due to the subject’s better performance in position learning. Different sensory threshold and recognition results all indicated that there is difference between amputees and able-bodied subjects. Similar interesting phenomena have also been observed. Both the amplitude and latency of the maximum ERP peaks for the amputee were smaller than those for the able-bodied subjects (Wang et al., 2022). One possible explanation is that the peripheral nerves regenerated in the stump were different from the intact one in structure and characteristics. Perhaps the nerve fibers in the residual stump may be fewer, and less sensitive compared to those in the intact limb. Another possible reason may be that the sensory neural pathways in amputee are different. For the amputee, electrical stimulation directly actives nerve endings of stump, arouses sensations, and transfers to the brain. But for able-bodied subjects, the electrical signals are transmitted to the nerves in the hand and returned back to the brain. In addition, brain reorganization after amputation (Chai et al., 2015; Björkman et al., 2016) may lead to different process of the central nerve system. The SRs of ES1, ES2, and ES3 were slightly lower than those of the remaining movement modes, which can be explained by the slightly lower SR (94.44%) of position sense for F60 and F30 than that for the other positions (100%). Seventy-five percent of FS1 (E60→E30→ST) positions were wrongly recognized as ES3 (ST→E60→E30) by amputee 2; these positions consist of two completely opposite movement modes. Because the subject’s recognition of the ST condition in the position sense experiment was low, the subject did not recognize the end position. Poor recognition of ST and F30 conditions was found in position sense experiments, resulting in poor recognition of ES2, ES3, and FS2 movements. We believe that for this subject, the confusion of a single position in this small-angle range movement misled the perception of the entire movement. When encoding wrist FE movements, F30, ST, and E30 using dual-channel stimulation may lead to confusion in the subjects’ perception.

For movement recognition, the DRT of able-bodied subjects was shorter than that of the two amputees, likely for the same reason mentioned for the wrist FE position sense. The DRTs of the two amputees did not show obvious differences in either successful recognition or wrong recognition. Both subjects were possibly confident that they had mastered the coding after their learning session for position sense and movement sense. In the learning session, all subjects were asked to watch the moving prosthetic limb or the wrist diagram on the display when they perceived the electrotactile stimulation. Other studies have shown that visual and tactile sensory systems share common features in object recognition, which proved that these systems have the potential to promote each other in the process of learning and cognition (Gonzalez et al., 2012; Tabrik et al., 2021). For transradial amputees, proprioception interruption caused by a missing wrist and hand decreased activity in their sensorimotor cortex circuit. After amputees underwent the above multisensory substitution training, their perceptual learning activity was induced in their sensorimotor cortex (Proulx et al., 2014). In our experiment, intuition guided the subjects to integrate the designed code method and prosthetic wrist movement without causing too much of a learning burden. The subjective responses of the two amputees to intuitiveness and ease of learning were in line with our expectations. Sensory feedback is important in the rehabilitation process of amputees who lack limb sensation, and the application of this system to somatosensory and motor training is expected to lead to enhanced motor and sensory cortical activation. The subjective responses of the two amputees to intuitiveness and ease of learning were in line with our expectations. Sensory feedback is important in the rehabilitation process of amputees who lack limb sensation, and the application of this system to somatosensory and motor training is expected to lead to enhanced motor and sensory cortical activation.

The current research had some limitations. First, too few amputees lead to the lack of universality of results and we will increase the number of amputees in future. Second, more evaluations, such as the time stability of the coding in combination with the forgetting curve (D’Anna et al., 2019), control performance of a prosthetic hand (Luo et al., 2021), and the activation of sensory cortex and mental burden through electroencephalogram (EEG) and functional near-infrared spectroscopy (fNIRS) (Midha et al., 2021; Zhu et al., 2021), need to be performed and the results need to be verified. Third, we focused on more reasonable experimental paradigms by setting the stimulation time and rest time (Buma et al., 2007; Marion et al., 2013).