Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Optimization of Semiautomated Calibration Algorithm of Multichannel Electrotactile Feedback for Myoelectric Hand Prosthesis

Optimization of Semiautomated Calibration Algorithm of Multichannel Electrotactile Feedback for... Hindawi Applied Bionics and Biomechanics Volume 2019, Article ID 9298758, 9 pages https://doi.org/10.1155/2019/9298758 Research Article Optimization of Semiautomated Calibration Algorithm of Multichannel Electrotactile Feedback for Myoelectric Hand Prosthesis 1,2 2,3 4 2 2 Milica Isaković , Jovana Malešević, Thierry Keller, Miloš Kostić, and Matija Štrbac University of Belgrade, School of Electrical Engineering, Belgrade, Serbia Tecnalia Serbia Ltd, Belgrade, Serbia University of Belgrade, Belgrade, Serbia Tecnalia Research & Innovation, San Sebastian, Spain Correspondence should be addressed to Milica Isaković; isakovic@etf.rs Received 15 December 2018; Accepted 10 February 2019; Published 14 March 2019 Academic Editor: Craig P. McGowan Copyright © 2019 Milica Isaković et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. The main drawback of the commercially available myoelectric hand prostheses is the absence of somatosensory feedback. We recently developed a feedback interface for multiple degrees of freedom myoelectric prosthesis that allows proprioceptive and sensory information (i.e., grasping force) to be transmitted to the wearer instantaneously. High information bandwidth is achieved through intelligent control of spatiotemporal distribution of electrical pulses over a custom-designed electrode array. As electrotactile sensations are location-dependent and the developed interface requires that electrical stimuli are perceived to be of the same intensity on all locations, a calibration procedure is of high importance. The aim of this study was to gain more insight into the calibration procedure and optimize this process by leveraging a priori knowledge. For this purpose, we conducted a study with 9 able-bodied subjects performing 10 sessions of the array electrode calibration. Based on the collected data, we optimized and simplified the calibration procedure by adapting the initial (baseline) amplitude values in the calibration algorithm. The results suggest there is an individual pattern of stimulation amplitudes across 16 electrode pads for each subject, which is not affected by the initial amplitudes. Moreover, the number of user actions performed and the time needed for the calibration procedure are significantly reduced by the proposed methodology. 1. Introduction recognize user intentions from electrical activity of remain- ing muscles (EMG) [3]. By employing the muscles origi- Humans rely on their hands to grasp, manipulate objects, nally used to accomplish the desired tasks, the user can and carry out a variety of activities of daily living. Tactile intuitively operate the artificial hand [4]. This holds a great feedback is one of the key components that enable dexter- promise to improving the quality of life for hand amputees, ous use of hands [1]. Hand amputation and loss of these which is why significant research efforts are aimed at essential functions are traumatic events leaving to dramatic further optimizing existing solutions [5] and providing a consequences on everyday life. Research and technological more intuitive user control [6]. However, the user will find advancement in prosthetic hands resulted in commercial a limited benefit from these improvements when using devices that today range from simple grippers with one multi-DOF prosthesis, if somatosensory feedback, which degree of freedom (DOF) to dexterous robotic hands that is crucial for effective motor planning and execution [7, 8] in grasping and object manipulation tasks [9], is missing. support multiple DOFs and grasping configurations [2]. The most technologically advanced noninvasive technique Including feedback and closing the loop in prosthetic to partially restore the functions of the missing hand is systems are important goals pursued by the researchers over by employing myoelectric prosthesis. These systems can the last four decades [10] and have also been acknowledged 2 Applied Bionics and Biomechanics In this paper, we present the results of research effort to from the user’s perspective as the most wanted improvement [11]. It needs to be noted that the advantages the feedback simplify the system personalisation and calibration process. provides are not limited to the aspect of control but that ben- First, the question of electrotactile variability is addressed in more detail to confirm the need for calibration and to efits can also be found in reduction of phantom-limb pain [12] and sensations of prosthesis embodiment [13]. determine possible strategies for algorithm simplification. Natural, somatotopically matched sensory feedback can Later, we investigate benefits of two different approaches be delivered to amputees invasively, via direct nerve [14– identified based on statistical analysis of the gathered data. 16] or brain [17] stimulation. Another approach to close the loop is known as sensory substitution [18]. In this 2. Materials and Methods method, the data are read from the prosthesis sensors and this information is transmitted to the user through a 2.1. System Setup. The system setup included a wireless controlled activation of his/her preserved sensory systems. multichannel electrotactile stimulation system (MAXSENS, The feedback can be delivered noninvasively through vibro- Tecnalia Research & Innovation, San Sebastian, ES) and a [19] or electrotactile [20] skin stimulation. In the latter, laptop PC (Intel® Core™ i5-4210U CPU at 1.70 GHz, low-level electrical current pulses are delivered to the skin 6 GB RAM) running MATLAB (R2016a, The MathWorks, to depolarize superficial afferents and elicit tactile sensation. Natick, MA) application with GUI for semiautomated The intensity and quality of sensation, and thereby the calibration of stimulation intensity. information content, can be regulated by changing the The stimulation system, presented in Figure 1, is a fully stimulation parameters (i.e., pulse width, amplitude, and/or programmable and integrated multichannel interface com- frequency coding) and the location of the stimulation prising a stimulation unit and a flexible array electrode delivery (i.e., spatial coding) [20]. [22]. The stimulation unit generates current-controlled Over the past decade, our group’s research efforts have biphasic stimulation pulses, with parameters suitable for been focused on leveraging the multipad electrode technol- the electrotactile stimulation (pulse width from 50 to ogy to deliver the high-quality proprioceptive and interac- 1000 μs with a 10 μs step, pulse rate from 1 to 400 Hz with tion force feedback in an intuitive manner [21–26]. The a 1 Hz step, and amplitude from 0.1 to 5 mA with a 0.1 mA developed approach relies on the dynamic stimulation pat- step). The unit is equipped with a Bluetooth communica- terns, where messages are coded using frequency and spatial tion interface, allowing control of the stimulation parame- modulation (i.e., changing the stimulation frequency and ters and active channels from the host PC using a simple location of the active electrode pad), to communicate the set of commands. The stimulation array electrode, with 16 state of the prosthesis in an intuitive manner [22]. Recently circular pads (cathodes) and a common adjacent anode, published results show that this approach enables precise was designed to be placed circumferentially around the communication of the prosthesis state [22] and that it has forearm. It was custom-designed and made on a 125 μm a steep learning curve, enabling intuitive use that does not PET substrate using Ag/AgCl conductive paste and an additionally burden the user [25]. insulation coating for biomedical applications covering the A drawback of electrotactile stimulation, amplified in conductive leads. The pads were covered with conductive multipad (array) systems, is the variability of elicited sensa- hydrogel (AG735, Axelgaard, DK) in order to improve the tions. Even though sensitivity to electrotactile stimulation contact between the electrode and the skin. has some topological regularity [21], when observed on the level of precision needed for sensory substitution, signif- icant intrasubject and temporal variability is observed [21, 2.2. Protocol. Nine able-bodied subjects (4 female, 5 male, 22, 27, 28]. An additional concern is the significant overlap age 29 ± 5 years, all right-handed) gave their informed of preferred and uncomfortable amplitude ranges between consent and participated in the study. The subjects were different subjects as described in [21]. This implies that an comfortably seated in front of the table with the laptop a priori set value that is in the preferred range for most users PC. The electrode was positioned circumferentially around could cause unpleasant sensations in some of them. To the subject’s left forearm, 5 cm below the elbow. It was overcome this and avoid any discomfort, a sequential positioned by ensuring that the two middle pads are cen- calibration procedure should be applied in every session. tered on the middle of the volar side (see Figure 1(a)). As described in our previous work [22–26], the method The electrode was positioned at the beginning and of limits [29] is an adequate procedure, where sequential removed at the end of each session. Each session was scanning of sensations in a predefined amplitude range is followed by at least 30 minutes of pause. Before the begin- iteratively performed for each pad until clear sensations of ning of the first session, each subject was introduced with similar intensity were observed throughout the electrode. an explanation of the calibration procedure and had the Over the course of our research, this procedure was stream- opportunity to familiarise with the GUI used. Subjects were lined to the point where it would rarely take more than 5 instructed to calibrate the stimulation intensity with a goal minutes, which was well within the acceptable range for the of obtaining similar sensations for all pads, sensations that experimental setup [22]. However, for everyday use of such are distinct, but pleasant, and to ensure that there is clear technology, this cumbersome setup procedure would be a spatial separation between the adjacent pads. An important strong deterrent for users and could significantly impede difference in respect to the previous studies [22–27] where the adoption. the calibration procedure was managed by the expert Applied Bionics and Biomechanics 3 from the results of previous calibration sessions to set a priori values for all pads. Here, following the logic out- lined in the discussion, we decided to set the baseline amplitudes of all pads at the 25th percentile value of all 90 standard calibration sessions. To test if in this way sim- ilar results can be obtained, on the fourth day each subject performed two additional sessions using this streamlined calibration procedure. 2.3. Data Analysis. Calibration curves obtained in all sessions (a) (b) of standard and streamlined calibration procedures were visually inspected. Figure 1: (a) The stimulation system, comprising the stimulation The coefficient of variation (CV), also known as relative unit and the array electrode inside the brace with an adjustable standard deviation (RSD), was calculated to examine the strap, positioned on the forearm. (b) The stimulation array dispersion of amplitudes for each subject and every pad. electrode with 16 circular cathodes and a common anode. It is expressed as a percentage and defined as the ratio of the standard deviation to the absolute mean value. researchers is that here the calibration was performed by In order to explore individual patterns of the curves the lay subjects without any assistance. which occur in all subjects, we applied correlation analysis. Each subject participated in 10 sessions of standard Correlations between each of the 10 calibration curves and calibration procedure, performed throughout 3 days. The their individual mean curve from 10 sessions, as well as the standard calibration procedure, which we previously used overall mean curve for all sessions and subjects, were calcu- in a study with able-bodied and amputee subjects [22], lated and averaged for each subject. The validity of the includes 2 phases. streamlined calibration procedure was confirmed through In phase 1, the PC application automatically increases correlation of the obtained calibration curve and the mean the stimulation amplitude of the first pad starting from curve from the standard procedure. For each subject, we also 1 mA with a 0.1 mA step until the subject indicates to have calculated the correlation between the baseline curve (overall perceived a pleasant, but distinct sensation by clicking the 25th percentile) and the final calibration curve in the appropriate button in GUI (Figure 2(a), STOP button). This streamlined calibration procedure. Paired-samples t-test is repeated for each of the 16 pads of the electrode. When was used to compare calculated correlation coefficients. phase 1 is finished, the subject is stimulated with every pad For each of the 90 selected amplitude curves (10 with the selected intensities in a fast scanning sequence. By sessions × 9 subjects), we calculated the total distance clicking the FAST button (Figure 2(b)), each pad is activated measured in mA, from 3 possible starting curves: for 0.2 seconds, starting from pad no. 1 and moving circum- ferentially to pad no. 16. This enables the subjects to quickly (1) Default constant used in the standard calibration feel transitions between the pads and test if the perceived process, i.e., 1 mA for each pad sensations are indeed similar for all pads. (2) Optimal constant, calculated as the value that results Phase 2 of the standard calibration procedure is aimed at in the smallest difference from all calibration curves, adjusting the baseline amplitudes obtained in phase 1 in a i.e., 1.8 mA simple and systematic manner. To allow subjects to identify subtle differences in sensations between adjacent pads and (3) Mean curve for all subjects and sessions fine-tune the amplitude, each pad is activated before and The total distance was calculated as the sum of absolute after the previous pad, as well as before and after the follow- differences between two values for all pads. Statistically ing pad. As an example, part of the sequence for fine-tuning significant differences between 3 baseline curves were pad no. 5 and no. 6 is 5-4-5-6-5-6-7-6. In this sequence, each assessed using a one-way repeated measure ANOVA with pad (except the first and last ones) is activated 3 times for 2 Greenhouse-Geisser correction, followed by a post hoc pair- seconds, so the whole fine-tuning process lasts 92 seconds. wise comparison with Bonferroni correction. The subject fine-tunes the intensities for each pad by clicking the “up” and “down” arrows on the corresponding slider (Figure 2(e)). At the end of the procedure, the subject is once 3. Results and Discussion again presented with calibrated intensities for all pads. Fine-tuning (phase 2) can be repeated if the subject is not To examine intersubject and intrasubject variability of pre- satisfied with the intensities, i.e., if the sensations of the same ferred stimulation amplitudes through the electrode array, intensity are not perceived for each pad of the electrode. we analysed the data of 10 standard calibration sessions After all subjects completed 10 sessions of standard cal- for individual subjects. The resulting preferred amplitudes ibration procedures, the protocol was modified to simplify obtained in this process are presented in Figure 3 for all 9 and accelerate the calibration procedure. Instead of starting subjects. Each panel contains stimulation amplitudes for from 1 mA and obtaining the baseline amplitudes through 16 pads obtained during 10 calibration sessions (coloured the phase 1 procedure, we leveraged the knowledge gained lines) and their mean value (black line). 4 Applied Bionics and Biomechanics Figure 2: Calibration GUI implemented in MATLAB. (a) STOP button used in phase 1 of the standard calibration procedure for stopping the increase in stimulation intensity when a satisfying sensation is reached. (b) FAST button for activation of fast scanning sequence of previously selected intensities. (c) FINE-TUNE button used to start the fine-tuning protocol. (d) NEXT button used for transition to the following pad if necessary. (e) “up” and “down” arrows for adjusting the amplitude of the corresponding pad during fine-tuning. 12 3 5 5 5 4 4 4 3 3 3 2 2 2 1 1 1 0 0 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 4 56 5 5 5 4 4 4 3 3 3 2 2 2 1 1 1 0 0 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 78 9 5 5 5 4 4 4 3 3 3 2 2 2 1 1 1 0 0 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 Pad number Figure 3: Amplitude values (vertical axis) for every electrode pad (horizontal axis) obtained through 10 sessions of calibration for each of 9 subjects. Individual measurements are presented through coloured lines, while the overall means are presented with a black line. The coefficient of variation (in %) for each subject and for existence of individual electrotactile maps and are also in the every pad is presented in Table 1. On average, the largest accordance with the findings described in [28], describing variance of 26.6% was noted in subject 5, while the smallest temporal variability of sensitivity to electrical stimulation. variance (13.1%) is in the measurements of the subject 8. To determine if a general rule can be deduced from these Despite the variance, distinctive individual patterns can results, we also analysed the group values. The correlations of be observed in visual inspection of the results presented in individual subject measurements with the overall mean curve Figure 3. were calculated, as shown in the first column of Table 2. Correlation coefficients between each measurement and These correlations are significantly smaller in compari- son with those calculated with respect to the individual aver- the mean for each subject were calculated to confirm the exis- tence of these patterns. As shown in the second column of age (paired-samples t-test, p =0 0033). However, for five of Table 2, the high correlation values confirm the existence of nine subjects they are larger than 0.65, which encouraged individual patterns. Such result concurs with [21] confirming us to further investigate the possibility of generalisation. Amplitude (mA) Applied Bionics and Biomechanics 5 Table 1: Coefficient of variation (in %) of selected amplitudes in the 10 initial sessions. Pad 123 45 678 9 10 11 12 13 14 15 16 Mean ± STD Subject 15 6±3 0 1 172020 1513 161710 151218 1313 151520 2 292018 1417 182323 192316 1628 14 8 15 18 8±5 4 21 7±6 8 3 15 26 20 25 27 31 22 12 13 7 28 23 30 21 23 24 18 1±5 7 4 191623 1817 202017 141617 1234 111124 26 6±7 3 5 203018 4429 303228 291717 2634 222030 22 4±8 2 6 23 25 12 19 23 31 23 20 28 8 28 16 18 44 19 21 15 6±5 1 7 251923 1623 101113 1512 9 1217 121319 8 17 14 11 10 12 14 15 11 9 10 13 18 11 15 14 16 13 1±2 7 15 4±4 5 9 181913 1514 121014 102021 1115 112618 Table 2: Comparison of correlation coefficients and the results of statistical analysis. Correlation between 10 Correlation between 10 Correlation between Correlation between calibration curves and mean calibration curves and streamlined calibration curve streamlined calibration Subject number curve from all sessions individual mean curve from 10 and individual mean curve curve and overall 25th (mean ± STD) sessions (mean ± STD) from 10 sessions percentile curve 0 86 ± 0 04 0 92 ± 0 05 1 0.90 0.88 0 35 ± 0 18 0 85 ± 0 10 2 0.68 0.23 0 69 ± 0 07 0 78 ± 0 10 3 0.84 0.69 0 76 ± 0 16 0 84 ± 0 10 4 0.93 0.90 0 24 ± 0 29 0 69 ± 0 15 5 0.87 0.05 0 47 ± 0 12 0 88 ± 0 05 6 0.83 0.52 0 40 ± 0 20 0 61 ± 0 15 7 0.66 0.74 8 0 74 ± 0 16 0 87 ± 0 11 0.95 0.83 0 74 ± 0 08 0 94 ± 0 06 9 0.96 0.85 Mean ± STD 0 59 ± 0 22 0 82 ± 0 11 0 84 ± 0 10 0 63 ± 0 31 Paired-samples p =0 0033 (significant) p =0 4502 (nonsignificant) p =0 0483 (significant) t-test A boxplot of selected stimulation amplitudes for all cal- shown in the first column of Table 2, the overall mean ibration sessions and all subjects (10 × 9) are presented in could present a good candidate for the a priori baseline. Figure 4. On visual inspection, this curve shows some of The second candidate we considered was the optimal the characteristic features noted in all individual patterns. constant, calculated as the value that results in the smallest The most obvious one is the convex shape, which again is difference from all selected amplitudes. in accordance with [21] and can be explained by the ana- To estimate the efficacy of the process depending on the tomical features. As shown in [21], the volar side of the baseline value, the amount of user actions needed to achieve forearm is more sensitive than the dorsal, and a similar the preferred stimulation amplitudes was calculated. This relation was noted between the medial and lateral sides. was defined as the cumulative “distance” in mA between the starting and selected amplitudes. The comparison was Furthermore, this may explain the peak intrasubject vari- ability (Table 1) observed in the pads that may shift from made between the two selected candidates, as well as with the dorsal to volar side based on the minor change in the the baseline used in the standard calibration process, which electrode placement, i.e., 4-5 medial and 13-14 lateral was 1 mA. The results are presented in Figure 5. (depending on the forearm size). A repeated measures ANOVA with a Greenhouse-Geisser correction determined that the mean difference between the In all calibration sessions, the time to reach the baseline was above three minutes, and the results were such that the selected values and the starting values differed statistically fine-tuning procedure was always required. This motivated significantly between 3 different options (F(1.038, 92 372 = us to determine a priori baseline values that are similar 113 639, p <0 0001). Post hoc tests using the Bonferroni enough to the desired result that it can be achieved through correction were also performed. The differences obtained the fine-tuning procedure only. As suggested by the results with optimal constant were significantly lower compared to 6 Applied Bionics and Biomechanics All subjects a procedure that seeks the minimal acceptable value is more appealing than the alternative. Since the overall data were 4.5 normally distributed (Anderson-Darling test, p <0 05 for all 16 pads), and therefore mean and median calibration 3.5 curves are very close, we decided to consider other descrip- 2.5 tive statistics such as quartiles and percentiles. When this additional constrain was applied, the 25th 1.5 percentile (i.e., lower quartile) curve became an obvious candidate for the baseline in the streamlined calibration 0.5 1 23456789 10 11 12 13 14 15 16 procedure. These values are within or below the range of Pad number selected amplitudes for each subject and are strongly correlated with the mean curve. Median As noted before, the streamlined procedure comprised 25th percentile only phase 2. Results of this calibration process are pre- Figure 4: Boxplots of amplitudes for 16 pads and all subjects. sented in Figure 6. We can note that the values obtained in Median value and 25th percentile of all data are presented with this way are within the range of the values obtained with red and green line, respectively. the standard calibration procedure and that the obtained curve follows a similar trend. This is confirmed by high values of correlation coefficients between the curve obtained 35 ⁎ in the optimized calibration session and the mean of 10 ses- sions, shown in Table 2. These range from 0.66 to 0.96, with an average of 0 84 ± 0 10. These are systematically higher than the correlation between the selected values and the baseline (25th percentile), shown in the last column of Table 2, suggesting that such starting point did not signifi- cantly influence the selection of amplitude values. All subjects managed to perform the calibration proce- 10 dure in a single sweep, meaning that the theoretical mini- mum of 92 seconds was reached. Furthermore, subjects 2, 5, and 6 noted that this procedure was lengthy, and in the second streamlined session opted not to use the fine-tuning procedure but several iterations of the “fast” mode (Figure 2(b)), where after quick scanning of all elec- trode pads they could correct the amplitudes where they detected an uneven sensation. As this was not foreseen in the experimental protocol and the time of execution was not measured, this cannot be reported in detail. However, it showcases the potential usability of multipad electrotac- tile technology, as well as of the proposed method. The study was conducted in able-bodied subjects, which Figure 5: Boxplots of distance from 3 different starting curves for all allowed unambiguous positioning of the electrode array by calibration curves. Horizontal bar with asterisks indicates measuring the distance from the elbow and ensuring that statistically significant difference in mean difference from the the two middle pads are centered on the middle of the volar calibration curves between the respective conditions. side of the forearm. It should be noted that the procedure the 1 mA starting line (p <0 0001). The tests revealed that the for electrode placement would be to a certain degree more overall mean starting curve resulted in smaller differences complex for amputees, since it must also consider some from the calibration curves (7 43 ± 2 75 mA), compared to additional factors like length of the stump, presence of neu- both the 1 mA constant (14 23 ± 6 04 mA) and the optimal roma, and problems with skin sensibility. However, based 1.8 mA constant (8 19 ± 2 80 mA), which were both statisti- on previous experience from several studies in patients with cally significant (p <0 0001). amputation [22, 23, 25, 26], small variations in electrode These results suggest that use of the overall mean as the position that were introduced due to these factors did not baseline value would present an efficient solution. However, affect calibration results or increase the time required for the values on the mean curve are above the range of selected electrode calibration. amplitudes on more than one pad for three subjects (4 pads The presented method of calibration was designed for for subject 3, 7 pads for subject 6, and 8 pads for subject 9). the electrotactile stimulation system with the 16-pad elec- Earlier experience [22–25] suggests that subjects prefer start- trode array positioned circumferentially around the fore- ing with lower intensity and reaching the desired value by arm, intended to be incorporated in the prosthesis socket increasing the intensity, than the other way around. Further- and present feedback from myoelectric prosthesis to the more, as higher amplitudes lead to faster habituation [19, 20], user. However, immediate results, i.e., calibration curves, Difference (mA) Amplitude (mA) 1 mA constant 1.8 mA constant (optimal) Overall mean Applied Bionics and Biomechanics 7 12 3 5 5 5 4 4 4 3 3 3 2 2 2 1 1 1 0 0 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 4 56 5 5 5 4 4 4 3 3 3 2 2 2 1 1 1 0 0 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 78 9 5 5 5 4 4 4 3 3 3 2 2 2 1 1 1 0 0 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 Pad number Figure 6: Results of the fine-tuning calibration (red line) compared to the results of the 10 calibration sessions (range— shaded area, mean—black line). are not relevant only for this single application, but can be knowledge in setting the baseline stimulation amplitudes, used as reference values when setting amplitudes of any type this process can be significantly streamlined, to a quarter of electrotactile display that is positioned on the forearm. As of initial time, without affecting the quality of the outcome. an example, these functions can be remapped to electrodes Based on these findings, we hypothesize that further improvements can be achieved by additionally leveraging intended to convey feedback information in teleoperation or virtual reality gaming. Moreover, an indirect result of this a posteriori knowledge in subsequent calibration sessions, study is the verification that adaptation of baseline in i.e., by choosing personalised baseline values informed by respect to electrode location can shorten and simplify the the previous selections by that specific subject. This will calibration procedure. Therefore, this same approach for be the subject of future investigation. identification of baseline values can be applied in electrotac- tile displays independent of electrode configuration and Data Availability position. These systems can be used for transmission of real- istic tactile sensation in many emerging application fields, The data used to support the findings of this study are such as providing feedback in lower-limb prostheses [30], available from the corresponding author upon request. vestibular substitution [31], assistive devices for the visually impaired applied to various body parts (back [32], tongue Conflicts of Interest [33], forehead [34], fingertips [35], or palm [36]), virtual reality and telexistence [37], and touch panels with haptic The authors declare that there is no conflict of interest feedback [38]. regarding the publication of this paper. 4. Conclusions Acknowledgments A novel method for calibration of the multipad electrotac- tile sensory substitution system was presented. The method The research was supported by Tecnalia Research & Inno- vation, Spain, and the Ministry of Education, Science and comprises a sequential algorithm supported by an intuitive GUI that allowed 9 lay subjects to perform the calibration Technological Development of Republic of Serbia (Project process in the time comparable to that reported when per- no. 175016). The authors would like to thank all the volun- formed by experts. Results show that with use of a priori teers who participated in this study. Amplitude (mA) 8 Applied Bionics and Biomechanics interface,” Proceedings of the National Academy of Sciences of References the United States of America, vol. 110, no. 45, pp. 18279– [1] C. L. MacKenzie and T. Iberall, The Grasping Hand (Vol. 104), 18284, 2013. Elsevier, 1994. [18] P. Bach-y-Rita and S. W. Kercel, “Sensory substitution and the [2] J. T. Belter, J. L. Segil, A. M. Dollar, and R. F. Weir, “Mechan- human–machine interface,” Trends in Cognitive Sciences, ical design and performance specifications of anthropomor- vol. 7, no. 12, pp. 541–546, 2003. phic prosthetic hands: a review,” Journal of Rehabilitation [19] K. A. Kaczmarek, J. G. Webster, P. Bach-y-Rita, and W. J. Research and Development, vol. 50, no. 5, pp. 599–618, 2013. Tompkins, “Electrotactile and vibrotactile displays for sensory [3] P. Parker, K. Englehart, and B. Hudgins, “Myoelectric signal substitution systems,” IEEE Transactions on Biomedical Engi- processing for control of powered limb prostheses,” Journal neering, vol. 38, no. 1, pp. 1–16, 1991. of Electromyography and Kinesiology, vol. 16, no. 6, pp. 541– [20] A. Y. J. Szeto and F. A. Saunders, “Electrocutaneous stimula- 548, 2006. tion for sensory communication in rehabilitation engineer- [4] M. A. Oskoei and H. Hu, “Myoelectric control systems—a sur- ing,” IEEE Transactions on Biomedical Engineering, vey,” Biomedical Signal Processing and Control, vol. 2, no. 4, vol. BME-29, no. 4, pp. 300–308, 1982. pp. 275–294, 2007. [21] M. Perović, M. Stevanović, T. Jevtić et al., “Electrical stimula- [5] S. Dosen, M. Markovic, N. Wille et al., “Building an internal tion of the forearm: a method for transmitting sensory signals model of a myoelectric prosthesis via closed-loop control for from the artificial hand to the brain,” Journal of Automatic consistent and routine grasping,” Experimental Brain Control, vol. 21, no. 1, pp. 13–18, 2013. Research, vol. 233, no. 6, pp. 1855–1865, 2015. [22] M. Štrbac, M. Belić, M. Isaković et al., “Integrated and flexible [6] N. Celadon, S. Došen, I. Binder, P. Ariano, and D. Farina, multichannel interface for electrotactile stimulation,” Journal “Proportional estimation of finger movements from of Neural Engineering, vol. 13, no. 4, article 046014, 2016. high-density surface electromyography,” Journal of Neuroen- [23] M. Isaković, M. Belić,M. Štrbac et al., “Electrotactile feedback gineering and Rehabilitation, vol. 13, no. 1, p. 73, 2016. improves performance and facilitates learning in the routine [7] J. Hermsdörfer, Z. Elias, J. D. Cole, B. M. Quaney, and D. A. grasping task,” European Journal of Translational Myology, Nowak, “Preserved and impaired aspects of feed-forward grip vol. 26, no. 3, 2016. force control after chronic somatosensory deafferentation,” [24] S. Dosen, M. Markovic, M. Strbac et al., “Multichannel electro- Neurorehabilitation and Neural Repair, vol. 22, no. 4, tactile feedback with spatial and mixed coding for closed-loop pp. 374–384, 2008. control of grasping force in hand prostheses,” IEEE Transac- [8] J. C. Rothwell, M. M. Traub, B. L. Day, J. A. Obeso, P. K. tions on Neural Systems and Rehabilitation Engineering, Thomas, and C. D. Marsden, “Manual motor performance in vol. 25, no. 3, pp. 183–195, 2017. a deafferented man,” Brain, vol. 105, no. 3, pp. 515–542, 1982. [25] M. Štrbac, M. Isaković, M. Belić et al., “Short- and long-term [9] R. S. Johansson and K. J. Cole, “Sensory-motor coordination learning of feedforward control of a myoelectric prosthesis during grasping and manipulative actions,” Current Opinion with sensory feedback by amputees,” IEEE Transactions on in Neurobiology, vol. 2, no. 6, pp. 815–823, 1992. Neural Systems and Rehabilitation Engineering, vol. 25, [10] D. S. Childress, “Closed-loop control in prosthetic systems: no. 11, pp. 2133–2145, 2017. historical perspective,” Annals of Biomedical Engineering, [26] M. Isaković,M. Štrbac, M. Belić et al., “Dynamic stimulation vol. 8, no. 4-6, pp. 293–303, 1980. patterns for conveying proprioceptive information from [11] B. Peerdeman, D. Boere, H. Witteveen et al., “Myoelectric fore- multi-DOF prosthesis,” in Converging Clinical and Engineer- arm prostheses: state of the art from a user-centered perspec- ing Research on Neurorehabilitation II. Biosystems & Biorobo- tive,” Journal of Rehabilitation Research & Development, tics, vol 15, J. Ibáñez, J. González-Vargas, J. Azorín, M. Akay, vol. 48, no. 6, pp. 719–737, 2011. and J. Pons, Eds., pp. 601–605, Springer, Cham, 2017. [12] C. Dietrich, K. Walter-Walsh, S. Preißler et al., “Sensory feed- [27] J. Malešević,M. Štrbac, M. Isaković et al., “Temporal and spa- back prosthesis reduces phantom limb pain: proof of a princi- tial variability of surface motor activation zones in hemiplegic ple,” Neuroscience Letters, vol. 507, no. 2, pp. 97–100, 2012. patients during functional electrical stimulation therapy ses- sions,” Artificial Organs, vol. 41, no. 11, pp. E166–E177, 2017. [13] C. Antfolk, M. D’Alonzo, B. Rosén, G. Lundborg, F. Sebelius, and C. Cipriani, “Sensory feedback in upper limb prosthetics,” [28] J. Malešević, S. Dedijer Dujović, A. M. Savić et al., “A decision Expert Review of Medical Devices, vol. 10, no. 1, pp. 45–54, support system for electrode shaping in multi-pad FES foot drop correction,” Journal of Neuroengineering and Rehabilita- tion, vol. 14, no. 1, p. 66, 2017. [14] D. Farina and O. Aszmann, “Bionic limbs: clinical reality and academic promises,” Science Translational Medicine, vol. 6, [29] G. A. Gescheider, Ch 3: the Classical Psychophysical Methods no. 257, article 257ps12, 2014. Psychophysics: the Fundamentals, Lawrence Erlbaum Associ- [15] S. Raspopovic, M. Capogrosso, F. M. Petrini et al., “Restoring ates, Mahwah, NJ, USA, 2013. natural sensory feedback in real-time bidirectional hand pros- [30] E. C. Wentink, E. J. Talsma-Kerkdijk, H. S. Rietman, and theses,” Science Translational Medicine, vol. 6, no. 222, article P. Veltink, “Feasibility of error-based electrotactile and audi- 222ra19, 2014. tive feedback in prosthetic walking,” Prosthetics and Orthotics [16] D. W. Tan, M. A. Schiefer, M. W. Keith, J. R. Anderson, International, vol. 39, no. 3, pp. 255–259, 2015. J. Tyler, and D. J. Tyler, “A neural interface provides [31] Y. P. Danilov, M. E. Tyler, K. L. Skinner, R. A. Hogle, and long-term stable natural touch perception,” Science Transla- P. Bach-y-Rita, “Efficacy of electrotactile vestibular substitu- tional Medicine, vol. 6, no. 257, article 257ra138, 2014. tion in patients with peripheral and central vestibular loss,” [17] G. A. Tabot, J. F. Dammann, J. A. Berg et al., “Restoring the Journal of Vestibular Research, vol. 17, no. 2-3, pp. 119–130, sense of touch with a prosthetic hand through a brain 2007. Applied Bionics and Biomechanics 9 [32] C. C. Collins, “Tactile television-mechanical and electrical image projection,” IEEE Transactions on Man-Machine Sys- tems, vol. 11, no. 1, pp. 65–71, 1970. [33] P. Bach-y-Rita, K. A. Kaczmarek, M. E. Tyler, and J. Garcia-Lara, “Form perception with a 49-point electrotactile stimulus array on the tongue: a technical note,” Journal of Rehabilitation Research and Development, vol. 35, no. 4, pp. 427–430, 1998. [34] H. Kajimoto, Y. Kanno, and S. Tachi, “Forehead electro-tactile display for vision substitution,” in Proc. EuroHaptics, Paris, France, 2006. [35] H. Kajimoto, N. Kawakami, S. Tachi, and M. Inami, “Smart- touch: electric skin to touch the untouchable,” IEEE Computer Graphics and Applications, vol. 24, no. 1, pp. 36–43, 2004. [36] H. Kajimoto, M. Suzuki, and Y. Kanno, “HamsaTouch: tactile vision substitution with smartphone and electro-tactile dis- play,” in CHI'14 Extended Abstracts on Human Factors in Computing Systems, pp. 1273–1278, New York, NY, USA, April-May 2014. [37] H. Kajimoto, “Design of cylindrical whole-hand haptic inter- face using electrocutaneous display,” in Haptics: Perception, Devices, Mobility, and Communication. EuroHaptics 2012. Lecture Notes in Computer Science, vol 7283, P. Isokoski and J. Springare, Eds., pp. 67–72, Springer, Berlin, Heidelberg, [38] H. Kajimoto, “Skeletouch: transparent electro-tactile display for mobile surfaces,” in SIGGRAPH Asia 2012 Emerging Tech- nologies, p. 21, ACM, New York, NY, USA, 2012. International Journal of Advances in Rotating Machinery Multimedia Journal of The Scientific Journal of Engineering World Journal Sensors Hindawi Hindawi Publishing Corporation Hindawi Hindawi Hindawi Hindawi www.hindawi.com Volume 2018 http://www www.hindawi.com .hindawi.com V Volume 2018 olume 2013 www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 Journal of Control Science and Engineering Advances in Civil Engineering Hindawi Hindawi www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 Submit your manuscripts at www.hindawi.com Journal of Journal of Electrical and Computer Robotics Engineering Hindawi Hindawi www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 VLSI Design Advances in OptoElectronics International Journal of Modelling & Aerospace International Journal of Simulation Navigation and in Engineering Engineering Observation Hindawi Hindawi Hindawi Hindawi Volume 2018 Volume 2018 Hindawi www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 www.hindawi.com www.hindawi.com www.hindawi.com Volume 2018 International Journal of Active and Passive International Journal of Antennas and Advances in Chemical Engineering Propagation Electronic Components Shock and Vibration Acoustics and Vibration Hindawi Hindawi Hindawi Hindawi Hindawi www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Applied Bionics and Biomechanics Hindawi Publishing Corporation

Optimization of Semiautomated Calibration Algorithm of Multichannel Electrotactile Feedback for Myoelectric Hand Prosthesis

Loading next page...
 
/lp/hindawi-publishing-corporation/optimization-of-semiautomated-calibration-algorithm-of-multichannel-k0242J5FFt
Publisher
Hindawi Publishing Corporation
Copyright
Copyright © 2019 Milica Isaković et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
ISSN
1176-2322
eISSN
1754-2103
DOI
10.1155/2019/9298758
Publisher site
See Article on Publisher Site

Abstract

Hindawi Applied Bionics and Biomechanics Volume 2019, Article ID 9298758, 9 pages https://doi.org/10.1155/2019/9298758 Research Article Optimization of Semiautomated Calibration Algorithm of Multichannel Electrotactile Feedback for Myoelectric Hand Prosthesis 1,2 2,3 4 2 2 Milica Isaković , Jovana Malešević, Thierry Keller, Miloš Kostić, and Matija Štrbac University of Belgrade, School of Electrical Engineering, Belgrade, Serbia Tecnalia Serbia Ltd, Belgrade, Serbia University of Belgrade, Belgrade, Serbia Tecnalia Research & Innovation, San Sebastian, Spain Correspondence should be addressed to Milica Isaković; isakovic@etf.rs Received 15 December 2018; Accepted 10 February 2019; Published 14 March 2019 Academic Editor: Craig P. McGowan Copyright © 2019 Milica Isaković et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. The main drawback of the commercially available myoelectric hand prostheses is the absence of somatosensory feedback. We recently developed a feedback interface for multiple degrees of freedom myoelectric prosthesis that allows proprioceptive and sensory information (i.e., grasping force) to be transmitted to the wearer instantaneously. High information bandwidth is achieved through intelligent control of spatiotemporal distribution of electrical pulses over a custom-designed electrode array. As electrotactile sensations are location-dependent and the developed interface requires that electrical stimuli are perceived to be of the same intensity on all locations, a calibration procedure is of high importance. The aim of this study was to gain more insight into the calibration procedure and optimize this process by leveraging a priori knowledge. For this purpose, we conducted a study with 9 able-bodied subjects performing 10 sessions of the array electrode calibration. Based on the collected data, we optimized and simplified the calibration procedure by adapting the initial (baseline) amplitude values in the calibration algorithm. The results suggest there is an individual pattern of stimulation amplitudes across 16 electrode pads for each subject, which is not affected by the initial amplitudes. Moreover, the number of user actions performed and the time needed for the calibration procedure are significantly reduced by the proposed methodology. 1. Introduction recognize user intentions from electrical activity of remain- ing muscles (EMG) [3]. By employing the muscles origi- Humans rely on their hands to grasp, manipulate objects, nally used to accomplish the desired tasks, the user can and carry out a variety of activities of daily living. Tactile intuitively operate the artificial hand [4]. This holds a great feedback is one of the key components that enable dexter- promise to improving the quality of life for hand amputees, ous use of hands [1]. Hand amputation and loss of these which is why significant research efforts are aimed at essential functions are traumatic events leaving to dramatic further optimizing existing solutions [5] and providing a consequences on everyday life. Research and technological more intuitive user control [6]. However, the user will find advancement in prosthetic hands resulted in commercial a limited benefit from these improvements when using devices that today range from simple grippers with one multi-DOF prosthesis, if somatosensory feedback, which degree of freedom (DOF) to dexterous robotic hands that is crucial for effective motor planning and execution [7, 8] in grasping and object manipulation tasks [9], is missing. support multiple DOFs and grasping configurations [2]. The most technologically advanced noninvasive technique Including feedback and closing the loop in prosthetic to partially restore the functions of the missing hand is systems are important goals pursued by the researchers over by employing myoelectric prosthesis. These systems can the last four decades [10] and have also been acknowledged 2 Applied Bionics and Biomechanics In this paper, we present the results of research effort to from the user’s perspective as the most wanted improvement [11]. It needs to be noted that the advantages the feedback simplify the system personalisation and calibration process. provides are not limited to the aspect of control but that ben- First, the question of electrotactile variability is addressed in more detail to confirm the need for calibration and to efits can also be found in reduction of phantom-limb pain [12] and sensations of prosthesis embodiment [13]. determine possible strategies for algorithm simplification. Natural, somatotopically matched sensory feedback can Later, we investigate benefits of two different approaches be delivered to amputees invasively, via direct nerve [14– identified based on statistical analysis of the gathered data. 16] or brain [17] stimulation. Another approach to close the loop is known as sensory substitution [18]. In this 2. Materials and Methods method, the data are read from the prosthesis sensors and this information is transmitted to the user through a 2.1. System Setup. The system setup included a wireless controlled activation of his/her preserved sensory systems. multichannel electrotactile stimulation system (MAXSENS, The feedback can be delivered noninvasively through vibro- Tecnalia Research & Innovation, San Sebastian, ES) and a [19] or electrotactile [20] skin stimulation. In the latter, laptop PC (Intel® Core™ i5-4210U CPU at 1.70 GHz, low-level electrical current pulses are delivered to the skin 6 GB RAM) running MATLAB (R2016a, The MathWorks, to depolarize superficial afferents and elicit tactile sensation. Natick, MA) application with GUI for semiautomated The intensity and quality of sensation, and thereby the calibration of stimulation intensity. information content, can be regulated by changing the The stimulation system, presented in Figure 1, is a fully stimulation parameters (i.e., pulse width, amplitude, and/or programmable and integrated multichannel interface com- frequency coding) and the location of the stimulation prising a stimulation unit and a flexible array electrode delivery (i.e., spatial coding) [20]. [22]. The stimulation unit generates current-controlled Over the past decade, our group’s research efforts have biphasic stimulation pulses, with parameters suitable for been focused on leveraging the multipad electrode technol- the electrotactile stimulation (pulse width from 50 to ogy to deliver the high-quality proprioceptive and interac- 1000 μs with a 10 μs step, pulse rate from 1 to 400 Hz with tion force feedback in an intuitive manner [21–26]. The a 1 Hz step, and amplitude from 0.1 to 5 mA with a 0.1 mA developed approach relies on the dynamic stimulation pat- step). The unit is equipped with a Bluetooth communica- terns, where messages are coded using frequency and spatial tion interface, allowing control of the stimulation parame- modulation (i.e., changing the stimulation frequency and ters and active channels from the host PC using a simple location of the active electrode pad), to communicate the set of commands. The stimulation array electrode, with 16 state of the prosthesis in an intuitive manner [22]. Recently circular pads (cathodes) and a common adjacent anode, published results show that this approach enables precise was designed to be placed circumferentially around the communication of the prosthesis state [22] and that it has forearm. It was custom-designed and made on a 125 μm a steep learning curve, enabling intuitive use that does not PET substrate using Ag/AgCl conductive paste and an additionally burden the user [25]. insulation coating for biomedical applications covering the A drawback of electrotactile stimulation, amplified in conductive leads. The pads were covered with conductive multipad (array) systems, is the variability of elicited sensa- hydrogel (AG735, Axelgaard, DK) in order to improve the tions. Even though sensitivity to electrotactile stimulation contact between the electrode and the skin. has some topological regularity [21], when observed on the level of precision needed for sensory substitution, signif- icant intrasubject and temporal variability is observed [21, 2.2. Protocol. Nine able-bodied subjects (4 female, 5 male, 22, 27, 28]. An additional concern is the significant overlap age 29 ± 5 years, all right-handed) gave their informed of preferred and uncomfortable amplitude ranges between consent and participated in the study. The subjects were different subjects as described in [21]. This implies that an comfortably seated in front of the table with the laptop a priori set value that is in the preferred range for most users PC. The electrode was positioned circumferentially around could cause unpleasant sensations in some of them. To the subject’s left forearm, 5 cm below the elbow. It was overcome this and avoid any discomfort, a sequential positioned by ensuring that the two middle pads are cen- calibration procedure should be applied in every session. tered on the middle of the volar side (see Figure 1(a)). As described in our previous work [22–26], the method The electrode was positioned at the beginning and of limits [29] is an adequate procedure, where sequential removed at the end of each session. Each session was scanning of sensations in a predefined amplitude range is followed by at least 30 minutes of pause. Before the begin- iteratively performed for each pad until clear sensations of ning of the first session, each subject was introduced with similar intensity were observed throughout the electrode. an explanation of the calibration procedure and had the Over the course of our research, this procedure was stream- opportunity to familiarise with the GUI used. Subjects were lined to the point where it would rarely take more than 5 instructed to calibrate the stimulation intensity with a goal minutes, which was well within the acceptable range for the of obtaining similar sensations for all pads, sensations that experimental setup [22]. However, for everyday use of such are distinct, but pleasant, and to ensure that there is clear technology, this cumbersome setup procedure would be a spatial separation between the adjacent pads. An important strong deterrent for users and could significantly impede difference in respect to the previous studies [22–27] where the adoption. the calibration procedure was managed by the expert Applied Bionics and Biomechanics 3 from the results of previous calibration sessions to set a priori values for all pads. Here, following the logic out- lined in the discussion, we decided to set the baseline amplitudes of all pads at the 25th percentile value of all 90 standard calibration sessions. To test if in this way sim- ilar results can be obtained, on the fourth day each subject performed two additional sessions using this streamlined calibration procedure. 2.3. Data Analysis. Calibration curves obtained in all sessions (a) (b) of standard and streamlined calibration procedures were visually inspected. Figure 1: (a) The stimulation system, comprising the stimulation The coefficient of variation (CV), also known as relative unit and the array electrode inside the brace with an adjustable standard deviation (RSD), was calculated to examine the strap, positioned on the forearm. (b) The stimulation array dispersion of amplitudes for each subject and every pad. electrode with 16 circular cathodes and a common anode. It is expressed as a percentage and defined as the ratio of the standard deviation to the absolute mean value. researchers is that here the calibration was performed by In order to explore individual patterns of the curves the lay subjects without any assistance. which occur in all subjects, we applied correlation analysis. Each subject participated in 10 sessions of standard Correlations between each of the 10 calibration curves and calibration procedure, performed throughout 3 days. The their individual mean curve from 10 sessions, as well as the standard calibration procedure, which we previously used overall mean curve for all sessions and subjects, were calcu- in a study with able-bodied and amputee subjects [22], lated and averaged for each subject. The validity of the includes 2 phases. streamlined calibration procedure was confirmed through In phase 1, the PC application automatically increases correlation of the obtained calibration curve and the mean the stimulation amplitude of the first pad starting from curve from the standard procedure. For each subject, we also 1 mA with a 0.1 mA step until the subject indicates to have calculated the correlation between the baseline curve (overall perceived a pleasant, but distinct sensation by clicking the 25th percentile) and the final calibration curve in the appropriate button in GUI (Figure 2(a), STOP button). This streamlined calibration procedure. Paired-samples t-test is repeated for each of the 16 pads of the electrode. When was used to compare calculated correlation coefficients. phase 1 is finished, the subject is stimulated with every pad For each of the 90 selected amplitude curves (10 with the selected intensities in a fast scanning sequence. By sessions × 9 subjects), we calculated the total distance clicking the FAST button (Figure 2(b)), each pad is activated measured in mA, from 3 possible starting curves: for 0.2 seconds, starting from pad no. 1 and moving circum- ferentially to pad no. 16. This enables the subjects to quickly (1) Default constant used in the standard calibration feel transitions between the pads and test if the perceived process, i.e., 1 mA for each pad sensations are indeed similar for all pads. (2) Optimal constant, calculated as the value that results Phase 2 of the standard calibration procedure is aimed at in the smallest difference from all calibration curves, adjusting the baseline amplitudes obtained in phase 1 in a i.e., 1.8 mA simple and systematic manner. To allow subjects to identify subtle differences in sensations between adjacent pads and (3) Mean curve for all subjects and sessions fine-tune the amplitude, each pad is activated before and The total distance was calculated as the sum of absolute after the previous pad, as well as before and after the follow- differences between two values for all pads. Statistically ing pad. As an example, part of the sequence for fine-tuning significant differences between 3 baseline curves were pad no. 5 and no. 6 is 5-4-5-6-5-6-7-6. In this sequence, each assessed using a one-way repeated measure ANOVA with pad (except the first and last ones) is activated 3 times for 2 Greenhouse-Geisser correction, followed by a post hoc pair- seconds, so the whole fine-tuning process lasts 92 seconds. wise comparison with Bonferroni correction. The subject fine-tunes the intensities for each pad by clicking the “up” and “down” arrows on the corresponding slider (Figure 2(e)). At the end of the procedure, the subject is once 3. Results and Discussion again presented with calibrated intensities for all pads. Fine-tuning (phase 2) can be repeated if the subject is not To examine intersubject and intrasubject variability of pre- satisfied with the intensities, i.e., if the sensations of the same ferred stimulation amplitudes through the electrode array, intensity are not perceived for each pad of the electrode. we analysed the data of 10 standard calibration sessions After all subjects completed 10 sessions of standard cal- for individual subjects. The resulting preferred amplitudes ibration procedures, the protocol was modified to simplify obtained in this process are presented in Figure 3 for all 9 and accelerate the calibration procedure. Instead of starting subjects. Each panel contains stimulation amplitudes for from 1 mA and obtaining the baseline amplitudes through 16 pads obtained during 10 calibration sessions (coloured the phase 1 procedure, we leveraged the knowledge gained lines) and their mean value (black line). 4 Applied Bionics and Biomechanics Figure 2: Calibration GUI implemented in MATLAB. (a) STOP button used in phase 1 of the standard calibration procedure for stopping the increase in stimulation intensity when a satisfying sensation is reached. (b) FAST button for activation of fast scanning sequence of previously selected intensities. (c) FINE-TUNE button used to start the fine-tuning protocol. (d) NEXT button used for transition to the following pad if necessary. (e) “up” and “down” arrows for adjusting the amplitude of the corresponding pad during fine-tuning. 12 3 5 5 5 4 4 4 3 3 3 2 2 2 1 1 1 0 0 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 4 56 5 5 5 4 4 4 3 3 3 2 2 2 1 1 1 0 0 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 78 9 5 5 5 4 4 4 3 3 3 2 2 2 1 1 1 0 0 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 Pad number Figure 3: Amplitude values (vertical axis) for every electrode pad (horizontal axis) obtained through 10 sessions of calibration for each of 9 subjects. Individual measurements are presented through coloured lines, while the overall means are presented with a black line. The coefficient of variation (in %) for each subject and for existence of individual electrotactile maps and are also in the every pad is presented in Table 1. On average, the largest accordance with the findings described in [28], describing variance of 26.6% was noted in subject 5, while the smallest temporal variability of sensitivity to electrical stimulation. variance (13.1%) is in the measurements of the subject 8. To determine if a general rule can be deduced from these Despite the variance, distinctive individual patterns can results, we also analysed the group values. The correlations of be observed in visual inspection of the results presented in individual subject measurements with the overall mean curve Figure 3. were calculated, as shown in the first column of Table 2. Correlation coefficients between each measurement and These correlations are significantly smaller in compari- son with those calculated with respect to the individual aver- the mean for each subject were calculated to confirm the exis- tence of these patterns. As shown in the second column of age (paired-samples t-test, p =0 0033). However, for five of Table 2, the high correlation values confirm the existence of nine subjects they are larger than 0.65, which encouraged individual patterns. Such result concurs with [21] confirming us to further investigate the possibility of generalisation. Amplitude (mA) Applied Bionics and Biomechanics 5 Table 1: Coefficient of variation (in %) of selected amplitudes in the 10 initial sessions. Pad 123 45 678 9 10 11 12 13 14 15 16 Mean ± STD Subject 15 6±3 0 1 172020 1513 161710 151218 1313 151520 2 292018 1417 182323 192316 1628 14 8 15 18 8±5 4 21 7±6 8 3 15 26 20 25 27 31 22 12 13 7 28 23 30 21 23 24 18 1±5 7 4 191623 1817 202017 141617 1234 111124 26 6±7 3 5 203018 4429 303228 291717 2634 222030 22 4±8 2 6 23 25 12 19 23 31 23 20 28 8 28 16 18 44 19 21 15 6±5 1 7 251923 1623 101113 1512 9 1217 121319 8 17 14 11 10 12 14 15 11 9 10 13 18 11 15 14 16 13 1±2 7 15 4±4 5 9 181913 1514 121014 102021 1115 112618 Table 2: Comparison of correlation coefficients and the results of statistical analysis. Correlation between 10 Correlation between 10 Correlation between Correlation between calibration curves and mean calibration curves and streamlined calibration curve streamlined calibration Subject number curve from all sessions individual mean curve from 10 and individual mean curve curve and overall 25th (mean ± STD) sessions (mean ± STD) from 10 sessions percentile curve 0 86 ± 0 04 0 92 ± 0 05 1 0.90 0.88 0 35 ± 0 18 0 85 ± 0 10 2 0.68 0.23 0 69 ± 0 07 0 78 ± 0 10 3 0.84 0.69 0 76 ± 0 16 0 84 ± 0 10 4 0.93 0.90 0 24 ± 0 29 0 69 ± 0 15 5 0.87 0.05 0 47 ± 0 12 0 88 ± 0 05 6 0.83 0.52 0 40 ± 0 20 0 61 ± 0 15 7 0.66 0.74 8 0 74 ± 0 16 0 87 ± 0 11 0.95 0.83 0 74 ± 0 08 0 94 ± 0 06 9 0.96 0.85 Mean ± STD 0 59 ± 0 22 0 82 ± 0 11 0 84 ± 0 10 0 63 ± 0 31 Paired-samples p =0 0033 (significant) p =0 4502 (nonsignificant) p =0 0483 (significant) t-test A boxplot of selected stimulation amplitudes for all cal- shown in the first column of Table 2, the overall mean ibration sessions and all subjects (10 × 9) are presented in could present a good candidate for the a priori baseline. Figure 4. On visual inspection, this curve shows some of The second candidate we considered was the optimal the characteristic features noted in all individual patterns. constant, calculated as the value that results in the smallest The most obvious one is the convex shape, which again is difference from all selected amplitudes. in accordance with [21] and can be explained by the ana- To estimate the efficacy of the process depending on the tomical features. As shown in [21], the volar side of the baseline value, the amount of user actions needed to achieve forearm is more sensitive than the dorsal, and a similar the preferred stimulation amplitudes was calculated. This relation was noted between the medial and lateral sides. was defined as the cumulative “distance” in mA between the starting and selected amplitudes. The comparison was Furthermore, this may explain the peak intrasubject vari- ability (Table 1) observed in the pads that may shift from made between the two selected candidates, as well as with the dorsal to volar side based on the minor change in the the baseline used in the standard calibration process, which electrode placement, i.e., 4-5 medial and 13-14 lateral was 1 mA. The results are presented in Figure 5. (depending on the forearm size). A repeated measures ANOVA with a Greenhouse-Geisser correction determined that the mean difference between the In all calibration sessions, the time to reach the baseline was above three minutes, and the results were such that the selected values and the starting values differed statistically fine-tuning procedure was always required. This motivated significantly between 3 different options (F(1.038, 92 372 = us to determine a priori baseline values that are similar 113 639, p <0 0001). Post hoc tests using the Bonferroni enough to the desired result that it can be achieved through correction were also performed. The differences obtained the fine-tuning procedure only. As suggested by the results with optimal constant were significantly lower compared to 6 Applied Bionics and Biomechanics All subjects a procedure that seeks the minimal acceptable value is more appealing than the alternative. Since the overall data were 4.5 normally distributed (Anderson-Darling test, p <0 05 for all 16 pads), and therefore mean and median calibration 3.5 curves are very close, we decided to consider other descrip- 2.5 tive statistics such as quartiles and percentiles. When this additional constrain was applied, the 25th 1.5 percentile (i.e., lower quartile) curve became an obvious candidate for the baseline in the streamlined calibration 0.5 1 23456789 10 11 12 13 14 15 16 procedure. These values are within or below the range of Pad number selected amplitudes for each subject and are strongly correlated with the mean curve. Median As noted before, the streamlined procedure comprised 25th percentile only phase 2. Results of this calibration process are pre- Figure 4: Boxplots of amplitudes for 16 pads and all subjects. sented in Figure 6. We can note that the values obtained in Median value and 25th percentile of all data are presented with this way are within the range of the values obtained with red and green line, respectively. the standard calibration procedure and that the obtained curve follows a similar trend. This is confirmed by high values of correlation coefficients between the curve obtained 35 ⁎ in the optimized calibration session and the mean of 10 ses- sions, shown in Table 2. These range from 0.66 to 0.96, with an average of 0 84 ± 0 10. These are systematically higher than the correlation between the selected values and the baseline (25th percentile), shown in the last column of Table 2, suggesting that such starting point did not signifi- cantly influence the selection of amplitude values. All subjects managed to perform the calibration proce- 10 dure in a single sweep, meaning that the theoretical mini- mum of 92 seconds was reached. Furthermore, subjects 2, 5, and 6 noted that this procedure was lengthy, and in the second streamlined session opted not to use the fine-tuning procedure but several iterations of the “fast” mode (Figure 2(b)), where after quick scanning of all elec- trode pads they could correct the amplitudes where they detected an uneven sensation. As this was not foreseen in the experimental protocol and the time of execution was not measured, this cannot be reported in detail. However, it showcases the potential usability of multipad electrotac- tile technology, as well as of the proposed method. The study was conducted in able-bodied subjects, which Figure 5: Boxplots of distance from 3 different starting curves for all allowed unambiguous positioning of the electrode array by calibration curves. Horizontal bar with asterisks indicates measuring the distance from the elbow and ensuring that statistically significant difference in mean difference from the the two middle pads are centered on the middle of the volar calibration curves between the respective conditions. side of the forearm. It should be noted that the procedure the 1 mA starting line (p <0 0001). The tests revealed that the for electrode placement would be to a certain degree more overall mean starting curve resulted in smaller differences complex for amputees, since it must also consider some from the calibration curves (7 43 ± 2 75 mA), compared to additional factors like length of the stump, presence of neu- both the 1 mA constant (14 23 ± 6 04 mA) and the optimal roma, and problems with skin sensibility. However, based 1.8 mA constant (8 19 ± 2 80 mA), which were both statisti- on previous experience from several studies in patients with cally significant (p <0 0001). amputation [22, 23, 25, 26], small variations in electrode These results suggest that use of the overall mean as the position that were introduced due to these factors did not baseline value would present an efficient solution. However, affect calibration results or increase the time required for the values on the mean curve are above the range of selected electrode calibration. amplitudes on more than one pad for three subjects (4 pads The presented method of calibration was designed for for subject 3, 7 pads for subject 6, and 8 pads for subject 9). the electrotactile stimulation system with the 16-pad elec- Earlier experience [22–25] suggests that subjects prefer start- trode array positioned circumferentially around the fore- ing with lower intensity and reaching the desired value by arm, intended to be incorporated in the prosthesis socket increasing the intensity, than the other way around. Further- and present feedback from myoelectric prosthesis to the more, as higher amplitudes lead to faster habituation [19, 20], user. However, immediate results, i.e., calibration curves, Difference (mA) Amplitude (mA) 1 mA constant 1.8 mA constant (optimal) Overall mean Applied Bionics and Biomechanics 7 12 3 5 5 5 4 4 4 3 3 3 2 2 2 1 1 1 0 0 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 4 56 5 5 5 4 4 4 3 3 3 2 2 2 1 1 1 0 0 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 78 9 5 5 5 4 4 4 3 3 3 2 2 2 1 1 1 0 0 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 Pad number Figure 6: Results of the fine-tuning calibration (red line) compared to the results of the 10 calibration sessions (range— shaded area, mean—black line). are not relevant only for this single application, but can be knowledge in setting the baseline stimulation amplitudes, used as reference values when setting amplitudes of any type this process can be significantly streamlined, to a quarter of electrotactile display that is positioned on the forearm. As of initial time, without affecting the quality of the outcome. an example, these functions can be remapped to electrodes Based on these findings, we hypothesize that further improvements can be achieved by additionally leveraging intended to convey feedback information in teleoperation or virtual reality gaming. Moreover, an indirect result of this a posteriori knowledge in subsequent calibration sessions, study is the verification that adaptation of baseline in i.e., by choosing personalised baseline values informed by respect to electrode location can shorten and simplify the the previous selections by that specific subject. This will calibration procedure. Therefore, this same approach for be the subject of future investigation. identification of baseline values can be applied in electrotac- tile displays independent of electrode configuration and Data Availability position. These systems can be used for transmission of real- istic tactile sensation in many emerging application fields, The data used to support the findings of this study are such as providing feedback in lower-limb prostheses [30], available from the corresponding author upon request. vestibular substitution [31], assistive devices for the visually impaired applied to various body parts (back [32], tongue Conflicts of Interest [33], forehead [34], fingertips [35], or palm [36]), virtual reality and telexistence [37], and touch panels with haptic The authors declare that there is no conflict of interest feedback [38]. regarding the publication of this paper. 4. Conclusions Acknowledgments A novel method for calibration of the multipad electrotac- tile sensory substitution system was presented. The method The research was supported by Tecnalia Research & Inno- vation, Spain, and the Ministry of Education, Science and comprises a sequential algorithm supported by an intuitive GUI that allowed 9 lay subjects to perform the calibration Technological Development of Republic of Serbia (Project process in the time comparable to that reported when per- no. 175016). The authors would like to thank all the volun- formed by experts. Results show that with use of a priori teers who participated in this study. Amplitude (mA) 8 Applied Bionics and Biomechanics interface,” Proceedings of the National Academy of Sciences of References the United States of America, vol. 110, no. 45, pp. 18279– [1] C. L. MacKenzie and T. Iberall, The Grasping Hand (Vol. 104), 18284, 2013. Elsevier, 1994. [18] P. Bach-y-Rita and S. W. Kercel, “Sensory substitution and the [2] J. T. Belter, J. L. Segil, A. M. Dollar, and R. F. Weir, “Mechan- human–machine interface,” Trends in Cognitive Sciences, ical design and performance specifications of anthropomor- vol. 7, no. 12, pp. 541–546, 2003. phic prosthetic hands: a review,” Journal of Rehabilitation [19] K. A. Kaczmarek, J. G. Webster, P. Bach-y-Rita, and W. J. Research and Development, vol. 50, no. 5, pp. 599–618, 2013. Tompkins, “Electrotactile and vibrotactile displays for sensory [3] P. Parker, K. Englehart, and B. Hudgins, “Myoelectric signal substitution systems,” IEEE Transactions on Biomedical Engi- processing for control of powered limb prostheses,” Journal neering, vol. 38, no. 1, pp. 1–16, 1991. of Electromyography and Kinesiology, vol. 16, no. 6, pp. 541– [20] A. Y. J. Szeto and F. A. Saunders, “Electrocutaneous stimula- 548, 2006. tion for sensory communication in rehabilitation engineer- [4] M. A. Oskoei and H. Hu, “Myoelectric control systems—a sur- ing,” IEEE Transactions on Biomedical Engineering, vey,” Biomedical Signal Processing and Control, vol. 2, no. 4, vol. BME-29, no. 4, pp. 300–308, 1982. pp. 275–294, 2007. [21] M. Perović, M. Stevanović, T. Jevtić et al., “Electrical stimula- [5] S. Dosen, M. Markovic, N. Wille et al., “Building an internal tion of the forearm: a method for transmitting sensory signals model of a myoelectric prosthesis via closed-loop control for from the artificial hand to the brain,” Journal of Automatic consistent and routine grasping,” Experimental Brain Control, vol. 21, no. 1, pp. 13–18, 2013. Research, vol. 233, no. 6, pp. 1855–1865, 2015. [22] M. Štrbac, M. Belić, M. Isaković et al., “Integrated and flexible [6] N. Celadon, S. Došen, I. Binder, P. Ariano, and D. Farina, multichannel interface for electrotactile stimulation,” Journal “Proportional estimation of finger movements from of Neural Engineering, vol. 13, no. 4, article 046014, 2016. high-density surface electromyography,” Journal of Neuroen- [23] M. Isaković, M. Belić,M. Štrbac et al., “Electrotactile feedback gineering and Rehabilitation, vol. 13, no. 1, p. 73, 2016. improves performance and facilitates learning in the routine [7] J. Hermsdörfer, Z. Elias, J. D. Cole, B. M. Quaney, and D. A. grasping task,” European Journal of Translational Myology, Nowak, “Preserved and impaired aspects of feed-forward grip vol. 26, no. 3, 2016. force control after chronic somatosensory deafferentation,” [24] S. Dosen, M. Markovic, M. Strbac et al., “Multichannel electro- Neurorehabilitation and Neural Repair, vol. 22, no. 4, tactile feedback with spatial and mixed coding for closed-loop pp. 374–384, 2008. control of grasping force in hand prostheses,” IEEE Transac- [8] J. C. Rothwell, M. M. Traub, B. L. Day, J. A. Obeso, P. K. tions on Neural Systems and Rehabilitation Engineering, Thomas, and C. D. Marsden, “Manual motor performance in vol. 25, no. 3, pp. 183–195, 2017. a deafferented man,” Brain, vol. 105, no. 3, pp. 515–542, 1982. [25] M. Štrbac, M. Isaković, M. Belić et al., “Short- and long-term [9] R. S. Johansson and K. J. Cole, “Sensory-motor coordination learning of feedforward control of a myoelectric prosthesis during grasping and manipulative actions,” Current Opinion with sensory feedback by amputees,” IEEE Transactions on in Neurobiology, vol. 2, no. 6, pp. 815–823, 1992. Neural Systems and Rehabilitation Engineering, vol. 25, [10] D. S. Childress, “Closed-loop control in prosthetic systems: no. 11, pp. 2133–2145, 2017. historical perspective,” Annals of Biomedical Engineering, [26] M. Isaković,M. Štrbac, M. Belić et al., “Dynamic stimulation vol. 8, no. 4-6, pp. 293–303, 1980. patterns for conveying proprioceptive information from [11] B. Peerdeman, D. Boere, H. Witteveen et al., “Myoelectric fore- multi-DOF prosthesis,” in Converging Clinical and Engineer- arm prostheses: state of the art from a user-centered perspec- ing Research on Neurorehabilitation II. Biosystems & Biorobo- tive,” Journal of Rehabilitation Research & Development, tics, vol 15, J. Ibáñez, J. González-Vargas, J. Azorín, M. Akay, vol. 48, no. 6, pp. 719–737, 2011. and J. Pons, Eds., pp. 601–605, Springer, Cham, 2017. [12] C. Dietrich, K. Walter-Walsh, S. Preißler et al., “Sensory feed- [27] J. Malešević,M. Štrbac, M. Isaković et al., “Temporal and spa- back prosthesis reduces phantom limb pain: proof of a princi- tial variability of surface motor activation zones in hemiplegic ple,” Neuroscience Letters, vol. 507, no. 2, pp. 97–100, 2012. patients during functional electrical stimulation therapy ses- sions,” Artificial Organs, vol. 41, no. 11, pp. E166–E177, 2017. [13] C. Antfolk, M. D’Alonzo, B. Rosén, G. Lundborg, F. Sebelius, and C. Cipriani, “Sensory feedback in upper limb prosthetics,” [28] J. Malešević, S. Dedijer Dujović, A. M. Savić et al., “A decision Expert Review of Medical Devices, vol. 10, no. 1, pp. 45–54, support system for electrode shaping in multi-pad FES foot drop correction,” Journal of Neuroengineering and Rehabilita- tion, vol. 14, no. 1, p. 66, 2017. [14] D. Farina and O. Aszmann, “Bionic limbs: clinical reality and academic promises,” Science Translational Medicine, vol. 6, [29] G. A. Gescheider, Ch 3: the Classical Psychophysical Methods no. 257, article 257ps12, 2014. Psychophysics: the Fundamentals, Lawrence Erlbaum Associ- [15] S. Raspopovic, M. Capogrosso, F. M. Petrini et al., “Restoring ates, Mahwah, NJ, USA, 2013. natural sensory feedback in real-time bidirectional hand pros- [30] E. C. Wentink, E. J. Talsma-Kerkdijk, H. S. Rietman, and theses,” Science Translational Medicine, vol. 6, no. 222, article P. Veltink, “Feasibility of error-based electrotactile and audi- 222ra19, 2014. tive feedback in prosthetic walking,” Prosthetics and Orthotics [16] D. W. Tan, M. A. Schiefer, M. W. Keith, J. R. Anderson, International, vol. 39, no. 3, pp. 255–259, 2015. J. Tyler, and D. J. Tyler, “A neural interface provides [31] Y. P. Danilov, M. E. Tyler, K. L. Skinner, R. A. Hogle, and long-term stable natural touch perception,” Science Transla- P. Bach-y-Rita, “Efficacy of electrotactile vestibular substitu- tional Medicine, vol. 6, no. 257, article 257ra138, 2014. tion in patients with peripheral and central vestibular loss,” [17] G. A. Tabot, J. F. Dammann, J. A. Berg et al., “Restoring the Journal of Vestibular Research, vol. 17, no. 2-3, pp. 119–130, sense of touch with a prosthetic hand through a brain 2007. Applied Bionics and Biomechanics 9 [32] C. C. Collins, “Tactile television-mechanical and electrical image projection,” IEEE Transactions on Man-Machine Sys- tems, vol. 11, no. 1, pp. 65–71, 1970. [33] P. Bach-y-Rita, K. A. Kaczmarek, M. E. Tyler, and J. Garcia-Lara, “Form perception with a 49-point electrotactile stimulus array on the tongue: a technical note,” Journal of Rehabilitation Research and Development, vol. 35, no. 4, pp. 427–430, 1998. [34] H. Kajimoto, Y. Kanno, and S. Tachi, “Forehead electro-tactile display for vision substitution,” in Proc. EuroHaptics, Paris, France, 2006. [35] H. Kajimoto, N. Kawakami, S. Tachi, and M. Inami, “Smart- touch: electric skin to touch the untouchable,” IEEE Computer Graphics and Applications, vol. 24, no. 1, pp. 36–43, 2004. [36] H. Kajimoto, M. Suzuki, and Y. Kanno, “HamsaTouch: tactile vision substitution with smartphone and electro-tactile dis- play,” in CHI'14 Extended Abstracts on Human Factors in Computing Systems, pp. 1273–1278, New York, NY, USA, April-May 2014. [37] H. Kajimoto, “Design of cylindrical whole-hand haptic inter- face using electrocutaneous display,” in Haptics: Perception, Devices, Mobility, and Communication. EuroHaptics 2012. Lecture Notes in Computer Science, vol 7283, P. Isokoski and J. Springare, Eds., pp. 67–72, Springer, Berlin, Heidelberg, [38] H. Kajimoto, “Skeletouch: transparent electro-tactile display for mobile surfaces,” in SIGGRAPH Asia 2012 Emerging Tech- nologies, p. 21, ACM, New York, NY, USA, 2012. International Journal of Advances in Rotating Machinery Multimedia Journal of The Scientific Journal of Engineering World Journal Sensors Hindawi Hindawi Publishing Corporation Hindawi Hindawi Hindawi Hindawi www.hindawi.com Volume 2018 http://www www.hindawi.com .hindawi.com V Volume 2018 olume 2013 www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 Journal of Control Science and Engineering Advances in Civil Engineering Hindawi Hindawi www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 Submit your manuscripts at www.hindawi.com Journal of Journal of Electrical and Computer Robotics Engineering Hindawi Hindawi www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 VLSI Design Advances in OptoElectronics International Journal of Modelling & Aerospace International Journal of Simulation Navigation and in Engineering Engineering Observation Hindawi Hindawi Hindawi Hindawi Volume 2018 Volume 2018 Hindawi www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 www.hindawi.com www.hindawi.com www.hindawi.com Volume 2018 International Journal of Active and Passive International Journal of Antennas and Advances in Chemical Engineering Propagation Electronic Components Shock and Vibration Acoustics and Vibration Hindawi Hindawi Hindawi Hindawi Hindawi www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 www.hindawi.com Volume 2018

Journal

Applied Bionics and BiomechanicsHindawi Publishing Corporation

Published: Mar 14, 2019

References