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

Learn More →

sEMG Signal Acquisition Strategy towards Hand FES Control

sEMG Signal Acquisition Strategy towards Hand FES Control Hindawi Journal of Healthcare Engineering Volume 2018, Article ID 2350834, 11 pages https://doi.org/10.1155/2018/2350834 Research Article 1 1 Cinthya Lourdes Toledo-Peral, Josefina Gutiérrez-Martínez , 1 2 Jorge Airy Mercado-Gutiérrez, Ana Isabel Martín-Vignon-Whaley, 3 3 Arturo Vera-Hernández , and Lorenzo Leija-Salas División de Investigación en Ingeniería Médica, Instituto Nacional de Rehabilitación “Luis Guillermo Ibarra Ibarra”, Calz. México- Xochimilco No. 289, Col. Arenal de Guadalupe, Tlalpan, 14389 Ciudad de México, Mexico Facultad de Ingeniería, Universidad La Salle, Benjamín Franklin 45, Col. Condesa, Cuauhtémoc, 06140 Ciudad de México, Mexico LAREMUS, Sección Bioelectrónica, Departamento de Ingeniería Eléctrica, Centro de Investigación y de Estudios Avanzados del Instituto Politécnico Nacional, Av. Instituto Politécnico Nacional 2508, Col. San Pedro Zacatenco, Gustavo A. Madero, 07360 Ciudad de México, Mexico Correspondence should be addressed to Josefina Gutiérrez-Martínez; josefina_gutierrez@hotmail.com Received 11 August 2017; Revised 1 December 2017; Accepted 27 December 2017; Published 14 March 2018 Academic Editor: Kunal Mitra Copyright © 2018 Cinthya Lourdes Toledo-Peral 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. Due to damage of the nervous system, patients experience impediments in their daily life: severe fatigue, tremor or impaired hand dexterity, hemiparesis, or hemiplegia. Surface electromyography (sEMG) signal analysis is used to identify motion; however, standardization of electrode placement and classification of sEMG patterns are major challenges. This paper describes a technique used to acquire sEMG signals for five hand motion patterns from six able-bodied subjects using an array of recording and stimulation electrodes placed on the forearm and its effects over functional electrical stimulation (FES) and volitional sEMG combinations, in order to eventually control a sEMG-driven FES neuroprosthesis for upper limb rehabilitation. A two-part protocol was performed. First, personalized templates to place eight sEMG bipolar channels were designed; with these data, a universal template, called forearm electrode set (FELT), was built. Second, volitional and evoked movements were recorded during FES application. 95% classification accuracy was achieved using two sessions per movement. With the FELT, it was possible to perform FES and sEMG recordings simultaneously. Also, it was possible to extract the volitional and evoked sEMG from the raw signal, which is highly important for closed-loop FES control. muscle activation, and their ability to span region is 1. Introduction curtailed [3]. Neurological disabilities are caused by damage of the ner- Biomedical signals, such as surface electromyography vous system (which includes the brain and spinal cord); this (sEMG), play a significant role in the measurement of the damage results in the loss of capacity to move and manipu- electrical muscle contraction. Plus, its analysis is one of late things, especially if fine movements are required [1]. The the standard procedures used to identify muscle actions effects of many neurological conditions can vary greatly in normal and pathologic conditions. sEMG signals can be from person to person, as well as from time to time for used for various applications, which include identifying neu- the same person. People with neurological conditions, such romuscular diseases, controlling signals for orthotic or pros- as a stroke, may present hand motor impairment and deficit thetic devices [4], anticipating movements of the muscles [5], controlling machines or robots, or detecting hand gestures to in motor execution, severe fatigue and/or weakness, impaired hand dexterity, tremors, spasticity, abnormal muscle syn- improve the quality of life [6]. ergies, and deficit in motor planning and motor learning sEMG patterns during movements exhibit a great deal of [2]. Stroke survivors may have great difficulty to modulate intersubject, intermuscle, and context-dependent variability. 2 Journal of Healthcare Engineering Understanding the sEMG interactions in hand movements is Skin cleaning a challenge [7]. Several researches have been directed to determine the extent to which each muscle participates in Electrodes template each synchronous and time-varying synergies for an individ- for sEMG location ualized human hand motor pattern [8] or to predict the sEMG patterns associated with static hand postures [9]. Recording and stimulation These studies show the importance of considering different electrodes placement intensities and durations of sEMG bursts, temporal patterns, strength of the muscle contraction [10], and muscle synergy Signal acquisition as a framework for sEMG patterns of hand postures. sEMG patterns are used for neuromuscular biofeedback OpenViBE [11], robot-aided [12] training, and neurorehabilitation, as Virtual platform well as to control devices such as neuroprosthesis based on Rest—10 s functional electrical stimulation (FES), to mimic a neuro- Hand open/power grasp/fine pinch/ muscular function for both upper and lower extremities pronation/supination —10 s [13], or to enhance hand motor recovery when physical ther- apy alone is ineffective in stroke patients [14] or with spinal 10 repetitions Functional cord injury [15]. electrical Several techniques have been employed for addressing stimulation Raw sEMG signal human hand movement patterns from sEMG signal. Tech- niques, such as an adaptive neuro-fuzzy inference system Signal processing integrated with a real-time learning scheme and time- frequency features, have been used to identify hand motion Figure 1: Electrode placement using a personalized template to find commands suitable for hand prosthesis control [16]. Ordinal sEMG signal for acquisition task and stimulation location. After pattern analysis is used to describe corrections of sEMG cleaning the skin and placing the electrodes, the isometric recordings during hand open and hand close states. The contraction (hand open, power grasp, fine pinch, pronation, and results suggest that the mutual information analysis has supination) was performed by the subject during 10 seconds, with potential in identifying different hand movements [17]. Usu- 10 seconds for rest. The task was repeated 10 times. A session included a task for each movement. ally, wavelet transformations and artificial neural network classifiers are used for hand movement analysis [10]. The Hilbert-Huang transform is another technique used to control of a neuroprosthesis to aid in motor neurorehabilita- detect, measure, filter, and decompose sEMG signals in tion of patients suffering from a stroke aftermath. order to identify patterns in time, frequency, or space or The presented technique is based on an array of recording and stimulation electrodes on the forearm, used the combination of flexion/extension arm movements. How- ever, the sEMG patterns can present abnormal muscle syn- to acquire sEMG signals from five hand motion patterns ergies and be indistinguishable [18]. This fact could make from six able-bodied subjects, and the effects of this tech- the classification in some stroke patients more difficult; for nique over functional electrical stimulation (FES) and voli- example, a solution proposed in [3] is to use voice recogni- tional sEMG combinations. tion as an auxiliary in a sEMG-driven actuated glove for clinical therapy purposes. 2. Methodology Recognizing sEMG signals with the aim of controlling assisting devices is not only concerned about feature extrac- 2.1. Identification of sEMG Locations. The first step was to tion and classification of signals but the acquisition site is also find the best electrode positioning for sEMG recording. of major importance. This position was found at the belly of the muscle, on the M-wave is an electrophysiological response evoked by upper part of the forearm, which is formed by the following electrostimulation detected in standard sEMG. It has been muscles: brachioradialis, palmaris longus, flexor carpi radia- studied widely in order to verify the functionality of the lis, flexor carpi ulnaris, extensor carpi radialis longus, and stimulation site measurement over the target muscle, extensor carpi ulnaris. Stimulation is performed at the ends which closely relates to muscle fiber recruitment. This of the same muscles. electrophysiologically driven approach is expected to lead In order to make sure that the electrodes were placed on to the identification of selective electrode configurations the same positions for the different trials for each subject, a of an array for functional movements [19]. However, find- personalized template was made. This template was created ing the best electrode configuration for sEMG recording to as follows: for bipolar channel placement, eight spots, where get the right sequence for movement activation still repre- the electrodes would be placed, were allocated and marked sents a challenge. on a piece of acetate paper. Then, the unique physical charac- This paper is related to the acquisition and analysis of teristics of the individual and the positions of five stimulation sEMG signals for active movements and to obtaining usable bipolar electrodes were marked on the same paper. Once the hand patterns with simultaneous placing of recording and places were allocated and the personalized template was stimulation electrodes on the forearm, for the eventual designed, sEMG acquisition was carried out. Journal of Healthcare Engineering 3 Clock stimulator Acquisition client Clock stimulator Clock stimulator Stimulation multiplexer Stream synchronization In|Out|Set Generic stream writer Signal display In|Out|Set In|Out|Set Display cue image Display cue image In|Out|Set In|Out|Set CSV file writer In|Out|Set Generic stream writer In|Out|Set (a) (b) Figure 2: OpenViBE flow diagram used to acquire raw sEMG signal (a); image cue synchronization control (b). This algorithm completes a movement task. Table 1: Stimulation electrode positions for each of the five target Raw sEMG signal movements. Target movement Electrode position Butterworth filter order 2.59-61 Hz Finger and wrist flexors. Flexor carpi Power grasp radialis, flexor carpi ulnaris, flexor digitorum superficialis. Application of Ulnar nerve. Flexor pollicis longus, flexor Daubechies Lumbrical grip digitorum superficialis. wavelet level-8 Finger and wrist extensors. Hand open Extensor carpi radialis. Extensor digitorum. sEMG signal–baseline Baseline Pronation Pronator teres. Supination Supinator. Application of Haar wavelet The subject’s skin was cleaned using an alcohol swab in level-8 order to reduce impedance and have a better coupling for the skin-electrode interface. Afterwards, the template was sEMG envelope placed on the subject’s forearm and marked; these were the spots where the electrodes should be placed. Figure 1 shows this procedure. The electrodes were kept in contact with the skin with a tubular mesh; this also reduced artifacts due to Figure 3: sEMG signal processing algorithm. The signal was filtered cable movements. for 60 Hz, baseline was subtracted through DWT, and the envelope Electrodes were connected to an open-source platform signal that selected the active pattern was obtained. called OpenViBE. This acquired the sEMG signal through a compatible open hardware acquisition device (OpenBCI) 2.2. sEMG Signals Acquisition. Six able-bodied subjects were which was connected to a designer space, where an algorithm included for the acquisition of sEMG signals, their age was designed for trial tasks (Figure 2). ranged from 21 to 33 years old, three males and three OpenViBE configuration was 24 for gain, 250 Hz for females. The subject was sitting in a comfortable position sampling rate, and eight channels for sEMG. The subject was asked to perform the movement shown in a cue image with his/her right arm supinated and leaning on the table. The subject was asked to perform an isometric contraction while it was on the screen. The task started with a rest for five movements: hand open, power grasp, fine pinch, of 10 seconds, and it continued with a ten-second isometric pronation, and supination. While contraction was active, contraction of hand open, power grasp, fine pinch, prona- the forearm muscles that participate in the motion were tion, or supination, depending on the trial. Cue images were shown alternatively until the subject completed ten palpated and located. 4 Journal of Healthcare Engineering Skin cleaning Functional Recording and stimulation electrical electrodes placement using FELT stimulation Signal acquisition OpenViBE OpenViBE OpenViBE Virtual platform Virtual platform Virtual platform Movement Contraction Rest Movement Contraction Rest Movement Contraction Rest Hand open 1 s 3 s Hand open 1 s 3 s Hand open 1 s 3 s Power grasp 1 s 3 s Power grasp 1 s 3 s Power grasp 1 s 3 s Fine pinch 1 s 3 s Fine pinch 1 s 3 s Fine pinch 1 s 3 s Pronation 1.8 s 2.2 s Pronation 1.8 s 2.2 s Pronation 1.8 s 2.2 s Supination 1.8 s 2.2 s Supination 1.8 s 2.2 s Supination 1.8 s 2.2 s × × × 5 repetitions 5 repetitions 5 repetitions Volitional sEMG signal Raw sEMG signal Evoked sEMG signal by FES + evoked by FES Signal processing Figure 4: sEMG signal acquisition for tasks (hand open, power grasp, fine pinch, pronation, and supination) with FES stimulation. An isometric contraction was performed by the subject for each part of the trial. The motion was repeated 5 times per part. A session included 5 repetitions of volitional contraction, followed by 5 repetitions of sEMG evoked by FES, and finally, 5 repetitions of volitional contraction plus the evoked sEMG by the FES stimulation. repetitions of the motion. A session was considered com- pleted when two movement tasks were finished (Figure 1). Table 2: Stimulation parameters for each subject and target All subjects completed two sessions for each of the men- movement. tioned motions. The tasks of sEMG recordings were saved Pulse Pulse Pulse as .csv files that included the information of eight chan- On/Off Subject Movement amplitude width frequency nels and a time vector. time (s) (mA ) (μ s) (Hz) From all the personalized positions, which were based on common regions found for each subject, a universal template PG 10 300 30 1/3 that kept the array for recording and stimulation electrodes LG 10 300 30 1/3 was designed. It was called forearm electrode set (FELT). 1 HO 10 500 30 1/3 SU 10 500 50 1.8/2.2 2.3. Preprocessing, Selection, and Feature Extraction. Each PR 10 500 30 1.8/2.2 sEMG record was imported into MATLAB® environment PG 10 300 30 1/3 for processing. From the .csv files, information of eight chan- LG 12 300 30 1/3 nels and a time vector was extracted. As seen in Figure 3, the signal was cleaned from line interference at 60 Hz by using a 2 HO 10 300 30 1/3 Butterworth filter, order 2, with a 59 to 61 Hz bandwidth. SU 10 300 50 1.8/2.2 After acquisition, data were conditioned using discrete PR 10 300 30 1.8/2.2 wavelet transforms (DWT). An eight-level decomposition PG 10 300 30 1/3 using mother wavelet Daubechies-4 was applied, and the LG 8 400 30 1/3 reconstructed signal was subtracted in order to eliminate 3 HO 12 300 30 1/3 baseline drift [20]; this was equivalent to filter a 0.7 Hz signal. Then, the DWT was applied, again, to an eight-level SU 10 500 50 1.8/2.2 decomposition, but this time a mother wavelet Haar was used PR 10 500 30 1.8/2.2 in order to find the envelope of the signal, which was PG: power grasp; HO: hand open; SU: forearm supination; PR: forearm obtained from its reconstruction. This envelope was used pronation; LG: lumbrical grip, applied through the RehaStim 2 electrical to find the parts of the sEMG signal that represented a stimulator. movement, in this case open hand or power grasp; then it was converted to a logic signal (Figure 3). Journal of Healthcare Engineering 5 ×10 Raw sEMG signal Processed sEMG signal 2.01 70 1.99 1.98 1.97 1.96 1.95 1.94 1.93 1.92 1.91 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 4 4 × 10 ×10 (a) (b) Figure 5: Subject 1, open hand/rest. Comparison of sEMG signal before and after processing using DWT. (a) Raw sEMG signal containing baseline drift and 60 Hz noise. (b) Processed sEMG signal drift-free and visible active and rest patterns. 500 500 0 0 ‒500 ‒500 0 20 40 60 80 100 120 140 160 180 200 0 20 40 60 80 100 120 140 160 180 200 CH1 CH2 (a) 0 0 ‒100 ‒100 ‒200 ‒200 0 20 40 60 80 100 120 140 160 180 200 0 20 40 60 80 100 120 140 160 180 200 CH1 CH2 (b) Figure 6: For subject 1, (a) open hand and (b) power grasp, sEMG processed and envelope signal obtained for active pattern selection. Example for channels 1 and 2 of 8. In order to find the characteristic features of the five target movements, the following parameters were calculated: SD = 〠 x − x , n − 1 mean absolute value—MAV (1), wave length—WL (2), zero i=1 crossing—ZC (3), standard deviation—SD (4), integral of absolute value—IAV (5), variance—V (6), and slope sign IAV = 〠 x , change—SSC (7). i=1 MAV = 〠 x , 1 i 2 V= 〠 x − x , 6 i=1 n − 1 i=1 WL = 〠 x − x , 2 1, x > x , x > x , i i−1 i i+1 i i−1 i=1 SSC = 7 1, x < x , x > x , i i+1 i i−1 1, x <0, x >0, i+1 i n−1 0, else ZC = 〠 1, x >0, x <0, 3 i+1 i i=1 From these parameters, a subset was selected for clas- 0, sification based on separability between movements and 6 Journal of Healthcare Engineering Mean absolute value Wavelength Standard deviation 60 350 70 50 300 60 40 250 50 30 200 40 20 150 30 10 100 20 0 50 10 −10 0 0 12345678 12345678 12345678 Channel Channel Channel Variance Integral of abs. value 4000 400 −1000 −2000 −100 12345678 12345678 Channel Channel Hand open Power grasp (a) ×10 Mean absolute value Wavelength Standard deviation 50 5 70 40 4 30 3 20 2 10 1 0 0 0 12345678 12345678 12345678 Channel Channel Channel ×10 Integral of abs. value Variance 5000 3.5 2.5 1.5 0.5 −1000 0 12345678 12345678 Channel Channel Hand open Power grasp (b) Figure 7: Analysis of window length for (a) 20 ms and (b) 3 s for all features (MAV, WL, SD, IAV, and V) and 8 channels, using data from the 6 subjects. Journal of Healthcare Engineering 7 classification accuracy. The new set of parameters were Table 3: Analysis of the combinations of selected channels and features with best performance during training, for each window used for classification. length. 2.4. Classification. For the classification of sEMG signals, Window Classifier Channels Features feature and window length analysis were performed for the length (s) accuracy (%) eight channels. The sEMG envelope signal was used for selec- 0.02 1–3 WL, SD 80.69 tion of active patterns at the processing stage. From this ~10 s 0.05 1–3 WL, SD 88.23 of sEMG activity, windows of 20 ms, 50 ms, 100 ms, 300 ms, 0.10 1–3 WL, SD, V 91.56 and 500 ms and 1 s and 3 s length, with a 25% overlap, were 0.30 1–3, 7-8 MAV, WL, SD, V 93.86 used to calculate the seven features described in (1), (2), (3), (4), (5), (6), and (7). 0.50 1–3, 7 MAV, WL, SD 95.83 A linear discriminant analysis (LDA) was executed for 1.00 1–4, 7-8 MAV, WL, SD, V, IAV 94.68 sets of two movements following the process described 3.00 1–3, 7-8 MAV, WL, SD, IAV 95.14 ahead, in this case for hand open and power grasp: The bold rows correspond to classification accuracies above 95%. (1) For each subject and each analysis window value, the the FELT was placed accordingly. For each target motion, seven features were extracted for the eight channels; there was a pair of self-adhesive stimulation electrodes for hand open task and power grasp task. (Axxelgard, USA) placed within the FELT, positioned as (2) The resulting 56 features obtained from each window presented in Table 1. were considered as a single trial for each movement. The subject was asked to perform an isometric contrac- tion for five movements: hand open, power grasp, fine pinch, (3) All available trials from the first session (one task pronation, and supination, but this time the trial consisted of per movement) of all subjects were concatenated three parts (Figure 4): movement-wise and randomized afterwards. (4) Label classes for each trial were set as 1 for hand open (1) Five isometric contractions of the selected move- and 2 for power grasp. ment, each lasting one second with three seconds rest (except pronation and supination: 1.8 active to 2.2 (5) For each window length value, the analysis was second rest) performed ten times. (2) Five FES stimulations of the selected movement, each (6) All trials were divided in 70% for a training set and lasting one second with three seconds rest 30% for a testing set. (3) Five isometric contractions during FES stimulations (7) A LDA classifier was trained with the training set. of the selected movement, each lasting one second (8) The trials on the testing set were classified with the with three seconds rest LDA classifier, and its classification accuracy was The algorithm in Figure 4 was performed once for each calculated as the ratio of correctly classified trials movement and subject. The stimulation parameters changed versus the total number of trials. for each movement according to Table 2. All subjects’ data from the first session (combinations of The new records were analyzed for processing the sEMG features, channels and window lengths) that obtained a clas- data because these signals included evoked and/or volitional sEMG as well as the FES stimulus. In order to extract the sification accuracy higher than 90% were chosen as the subset of features used to train the final LDA classifier. Data from sEMG evoked/volitional sEMG from the stimulus artifact, a comb-type filter was applied to eliminate the 30 or 50 Hz sig- the second session, which consisted of hand open and power grasp for each subject, was processed in the same way and nal of the stimulus, by means of a Butterworth filter, order was used to test the LDA classifier. two, with a 29 to 31 Hz or 49 to 51 Hz bandwidths, accord- ingly. All data processing is designed and performed in MATLAB environment. The parameters calculated for these 2.5. sEMG Recording and FES Application: Acquisition and signals are MAV (1) and root mean square (RMS) (8) to Processing. For the trials of sEMG signal acquisition during compare sEMG of evoked and volitional and evoked signals. FES application, the acquisition was performed using the OpenViBE platform and OpenBCI device with the same con- figuration mentioned above. For FES application, a RehaStim RMS = 〠 x 8 2 electrical stimulator (Hasomed Gmbh, Germany) was used i=1 and programmed in an interface developed in Simulink®/ MATLAB Environment. 3. Results and Discussion Three able-bodied subjects out of the six that performed the previous trials without FES, age range from 22 to 34 A personalized template was designed for each subject. years old, two males and one female, were included for These templates were used to successfully locate muscle sEMG acquisition. Their skin was cleaned with alcohol and sites and place electrodes for the second trial, with the 8 Journal of Healthcare Engineering 400 400 200 200 0 0 ‒200 ‒200 0 20 40 60 80 100 120 140 160 180 200 0 20 40 60 80 100 120 140 160 180 200 CH1 CH2 (a) 0 0 ‒500 ‒500 0 20 40 60 80 100 120 140 160 180 200 0 20 40 60 80 100 120 140 160 180 200 CH1 CH2 (b) Figure 8: Subject 1 using FELT: (a) channels 1 and 2 for open hand and (b) channels 1 and 2 for power grasp. BL/60 Hz ‒500 ‒1000 ‒1500 0 50 100 150 Time (s) sEMG (a) Spectrogram 1.5 ‒20 ‒40 ‒60 0.5 ‒80 ‒100 020 40 60 80 100 120 Frequency (Hz) (b) Figure 9: (a) Power grasp sEMG signal recorded from trial (algorithm Figure 4), channel 1. Baseline has been eliminated using algorithm of Figure 2. (b) Spectogram of sEMG signal, where activity in the 30 Hz band for the 2nd and 3rd sets of motions and their harmonics can be observed. advantage of a tenfold reduction in location time, approx- positions to acquire eight sEMG channels and enough imately. Then, the FELT was designed as a universal array place for five bipolar stimulation channels. It is important from all the individual templates. to mention that since the forearm is a small area, it was The main purpose of the FELT was to simplify recording difficult to find the right allocation for all the electrodes and stimulation electrode placing, for a future FES-based (stimulation electrodes are 5×5 cm and recording electrodes are 1 cm in diameter) and still have useful signals that could neuroprosthesis clinical application for stroke aftermath rehabilitation at upper limb and hand. There are not stan- be processed and classified. dardized designs for sEMG recording and FES application. Due to this critical disposition, the electrode locations The sEMG signals acquired for open hand and power grasp from the personalized templates were assessed through the were used to evaluate the right position of the recording elec- sEMG signals obtained by means of signal processing and classification of movements. trodes at the FELT. One of the objectives of this work was to allocate all A baseline drift-free signal was obtained from the raw electrodes keeping the balance between having available sEMG signal during the preprocessing stage (Figure 5). All Amplitude (mV) Time (min) Power/frequency (dB/Hz) Journal of Healthcare Engineering 9 Volitional sEMG Frequency spectrum Evoked sEMG without FES Frequency spectrum 600 30 800 70 400 600 25 60 ‒200 ‒200 30 ‒400 ‒400 ‒600 ‒600 10 ‒800 ‒800 0 ‒1000 0 30 35 40 45 50 55 0 20 40 60 80 100 120 140 70 75 80 85 90 95 0 20 40 60 80 100120140 Time (s) Frequency (Hz) Time (s) Frequency (Hz) ‒20 ‒40 ‒40 ‒80 ‒60 ‒120 16 20 14 ‒100 80 ‒80 60 10 ‒60 10 40 6 ‒40 20 5 2 ‒20 Frequency (Hz) 0 Time (s) Frequency (Hz) Time (s) (a) (b) Evoked and volitional Frequency spectrum sEMG without FES 200 30 ‒200 20 ‒400 ‒600 10 ‒800 5 ‒1000 0 20 40 60 80 100120 140 110 115 120 125 130 135 140 Frequency (Hz) Time (s) ‒20 ‒40 ‒60 80 15 0 Time (s) Frequency (Hz) (c) Figure 10: Power grasp, subject 1, channel 1, sEMG signals of the 3 parts of the trial. (a) Set of 5 isometric contractions of the selected movement, each lasting 1 second with 3 seconds rest. (b) 5 FES stimulations of the selected movement, each lasting 1 second with 3 seconds rest. (c) 5 isometric contractions during FES stimulations of the selected movement, each lasting 1 second with 3 seconds rest. Figure 6 shows an example of two of the eight sEMG sessions from the six subjects were put through this process- ing. sEMG signal in Figure 5(a) has a large baseline, while channels processed and the envelope signal obtained, which Figure 5(b) shows a cleaner sEMG signal despite original shows the active sEMG sections selected. These correspond baseline drifting; also, the differences between each contrac- to open hand and power grasp movements. tion repetition are clearer. From the analysis of the combinations of features, chan- The preprocessing analysis and processing method nel, and window length for all subjects, it was found that only showed that no matter the 60 Hz noise and drifting base- 5 features (MAV, WL, SD, IAV, and V) yield enough infor- line, the signal could be isolated for feature extraction and mation for classification, above 90% accuracy (Figure 7). In classification. It is important to mention that if the acquisi- Figure 7(b), it can be observed that when the length of the tion signal was less contaminated, this process could be fas- window was larger, for features like MAV or SD, it was easier ter and closer to real time for control applications, which to find a clear separation of the value of the parameters. Even emphasizes the need to design and build a specialized the smaller windows, i.e., 20 ms, (Figure 7(a)) performed with acquisition stage in order to start with the best version of an accuracy of 80.69%. Then, it is important to find a com- a raw sEMG signal (which can also consider a configura- promise between window length and classifier performance. tion that allows the simultaneous application of FES, for From this analysis, using 9 out of 10 repetitions of each volitional sEMG extraction). movement per session and considering session 1 for training Amplitude (mV) Amplitude Amplitude (mV) Amplitude (mV) Amplitude Amplitude 10 Journal of Healthcare Engineering sEMG signal evoked and FES application Volitional sEMG and FES application 500 500 0 0 ‒500 ‒500 ‒1000 ‒1000 ‒1500 ‒1500 72 74 76 78 80 82 84 86 88 90 92 110 115 120 125 130 135 140 Time (s) Time (s) sEMG evoked by FES Volitional sEMG and evoked by FES 800 800 400 400 0 0 ‒200 ‒200 ‒400 ‒400 ‒600 ‒600 ‒800 ‒800 ‒1000 ‒1000 72 74 76 78 80 82 84 86 88 90 92 110 115 120 125 130 135 140 Time (s) Time (s) (a) (b) Figure 11: Comparison of sEMG signals between 2 parts of the trial involving FES application. (a) Raw signal including FES (top) and sEMG signal evoked by FES free of the stimulus (bottom). (b) Raw signal including volitional sEMG and FES (top) and volitional sEMG signal and evoked by FES free of the stimulus (bottom). and session 2 for testing, it was found that MAV, WL, and SD Table 4: RMS and MAV values obtained for 3 able-bodied subjects, comparison between sEMG evoked by FES and the combination of features and a 0.50 seconds’ window length were the best volitional and evoked by FES signals. Values obtained from motion combinations for the classifier to perform with only 4 of power grasp, channel 1. channels (CH1, CH2, CH3, and CH7) at a 95.83% classifica- tion accuracy. The results from all combinations can be seen Volitional sEMG in Table 3. sEMG evoked by FES + sEMG evoked by FES Subject Gender This study and analysis was performed to minimize RMS MAV RMS MAV inputs for the classifier, with the aim of getting a closer (mV) (mV) (mV) (mV) approach to a real-time application. This analysis is a classi- 1 Male 147.5061 105.6109 147.4792 104.6412 fication method for multisubjects, used to generate a 2 Male 159.2150 109.0613 169.7005 126.7346 sEMG-driven control for a FES neuroprosthesis application. 3 Female 306.5072 200.5491 215.4075 138.3950 An example of the sEMG signals obtained for subject 1 using FELT, for channels 1 and 2, is shown in Figure 8. It can be observed that even though the signal was noisier for and five bipolar stimulation channels (larger electrodes, this session, the processing algorithm was still able to find 5×5 cm) in the forearm, which is a small area for so many the active sEMG sections. electrodes (a total of 27). Figure 9 shows the signal resulting from the sEMG Signal processing yielded a very clean signal that pre- (evoked/volitional) and FES stimulus signal acquisition using served sEMG components by using DWT and allowed to dif- ferentiate between movements through feature extraction the FELT. Figure 10 shows each set of repetitions of the 3 parts and classification. of the trial. The frequency spectrum and a 3D spectrogram We found an optimal combination between window are presented. length and number of channels and features, at 0.5 seconds, Figure 11 shows the sets of contractions for sEMG evoked with four channels and three features (MAV, WL, and SD), which allowed a more efficient classification in terms of time by FES and those from a volitional sEMG contribution used in order to compare the effects of both conditions. and channels. The RMS and MAV values for each repetition were The stimulation parameters were selected in order to calculated; Table 4 shows an example of these values. generate a complete movement without subject discomfort; however, range of movement is yet to be evaluated. As for signal processing, knowing the stimulus frequency before- 4. Conclusions hand allows the use of a filtering technique feasible for offline and online application. From Figure 10, it is evident that a The design of a personalized template presented in this paper replicates the sEMG signal between sessions. Also, natural sEMG contraction activates the slow fibers of the the forearm electrode set (FELT) resulted from the need muscle, but in the cases of FES application (Figures 10(b) and 10(c)), the fast twitch fibers have a larger contribution to find the correct place for eight sEMG bipolar channels Amplitude (mV) Amplitude (mV) Amplitude (mV) Amplitude (mV) Journal of Healthcare Engineering 11 [8] M. D. Klein Breteler, K. J. Simura, and M. Flanders, “Timing of to the sEMG record. Additionally, the evoked and voli- muscle activation in a hand movement sequence,” Cerebral tional sEMG with FES were similar; however, it should Cortex, vol. 17, no. 4, pp. 803–815, 2007. be considered that the sample was small and that all sub- [9] A. B. Ajibove and R. F. Weir, “Muscle synergies as a predictive jects were able-bodied. Therefore, a protocol with a bigger framework for the EMG patterns of new hand postures,” Jour- sample is needed and it still remains to be seen if these nal of Neural Engineering, vol. 6, no. 3, article 036004, 2009. results hold for patients. [10] S. M. Mane, R. A. Kambli, F. S. Kazi, and N. M. Singh, “Hand Using the FELT, it was possible to perform sEMG motion recognition from single channel surface EMG using recording and FES simultaneously. Moreover, it was possi- wavelet & artificial neural network,” Procedia Computer Sci- ble to extract the volitional and evoked sEMG from the ence, vol. 49, pp. 58–65, 2015. raw signal, which was accomplished without blanking the [11] O. M. Giggins, U. M. Persson, and B. Caulfield, “Biofeedback signal allowing better control techniques to be implemented. in rehabilitation,” Journal of Neuroengineering and Rehabilita- This is highly important for closed-loop FES control. tion, vol. 10, no. 1, p. 60, 2013. In the evoked/volitional sEMG and FES trials, the FES [12] B. Cesqui, P. Tropea, S. Micera, and H. Krebs, “EMG-based stimulus was successfully eliminated from the recorded sig- pattern recognition approach in post stroke robot-aided reha- nal leaving a usable sEMG signal for FES control and other bilitation: a feasibility study,” Journal of Neuroengineering and applications as orthosis, prosthetics, neuroprosthesis, and Rehabilitation, vol. 10, no. 1, p. 75, 2013. other rehabilitation and assistive devices. [13] D. B. Popović, T. Sinkaer, and M. B. Popović, “Electrical stim- ulation as a means for achieving recovery of function in stroke Conflicts of Interest patients,” NeuroRehabilitation, vol. 25, no. 1, pp. 45–58, 2009. The authors declare that they have no conflicts of interest. [14] F. Quandt and F. C. Hummel, “The influence of functional electrical stimulation on hand motor recovery in stroke patients: a review,” Experimental & Translational Stroke Med- Acknowledgments icine, vol. 6, no. 1, p. 9, 2014. The authors would like to thank the National Council for [15] C. H. Ho, R. J. Triolo, A. L. Elias et al., “Functional electri- cal stimulation and spinal cord injury,” Physical Medicine Science and Technology (Consejo Nacional de Ciencia y and Rehabilitation Clinics of North America, vol. 25, no. 3, Tecnología—CONACyT) for supporting the Project CON- pp. 631–654, 2014. ACYT-SALUD-2016-1-272983 as well as Gabriel Vega [16] M. Khezri and M. Jahed, “A novel approach to recognize hand Martínez, M.Sc., for his contributions. movements via sEMG patterns,” in 2007 29th Annual Interna- tional Conference of the IEEE Engineering in Medicine and References Biology Society, pp. 4907–4910, Lyon, France, August 2007. [1] S. W. Lee, K. M. Wilson, B. A. Lock, and D. G. Kamper, “Sub- [17] G. Ouyang, Z. Ju, and H. Liu, “Mutual information analy- ject-specific myoelectric pattern classification of functional sis with ordinal pattern for EMG based hand motion recogni- hand movements for stroke survivors,” IEEE Transactions on tion,” in ICIRA, Intelligent Robotics and Applications, vol. 7506 Neural Systems and Rehabilitation Engineering, vol. 19, no. 5, of Lecture Notes in Computer Science, pp. 499–506, Springer, pp. 558–566, 2011. Berlin, Heidelberg, 2012. [2] P. Raghavan, “The nature of hand motor impairment after [18] Y. Lan, J. Yao, and J. Dewald, “The impact of shoulder stroke and its treatment,” Current Treatment Options in abduction loading on EMG-based intention detection of hand Cardiovascular Medicine, vol. 9, no. 3, pp. 221–228, 2007. opening and closing after stroke,” in 2011 Annual Interna- tional Conference of the IEEE Engineering in Medicine and [3] D. G. Kamper, “Restoration of hand function in stroke and spinal cord injury,” in Neurorehabilitation Technology,D. Biology Society, pp. 4136–4139, Boston, MA, USA, September Reinkensmeyer and V. Dietz, Eds., pp. 311–324, Springer, Cham, 2nd edition, 2016. [19] C. De Marchis, T. S. Monteiro, C. Simon-Martinez, [4] A. Altamirano, A. Vera, R. Muñoz, L. Leija, and D. Wolf, S. Conforto, and A. Gharabaghi, “Multi-contact functional “Multichannel sEMG signal analysis using Hilbert-Huang electrical stimulation for hand opening: electrophysiologically transform to identify time-frequency features,” in 36th Annual driven identification of the optimal stimulation site,” Journal International Conference of the IEEE Engineering in Medicine of Neuroengineering and Rehabilitation, vol. 13, no. 1, p. 22, and Biology Society, Chicago, IL, USA, August 2014. [5] Y. Hou, J. Zurada, and W. Karwowski, “Prediction of EMG sig- [20] G. Vega-Martinez, C. Alvarado-Serrano, and L. Leija, “ECG nals of trunk muscles in manual lifting using a neural network baseline drift removal using discrete wavelet transform,” in model,” in 2004. Proceedings. 2004 IEEE International Joint 2011 8th International Conference on Electrical Engineering Conference on Neural Networks, pp. 1935–1940, Budapest, Computing Science and Automatic Control (CCE), Merida Hungary, January 2004. City, Mexico, October 2011. [6] E. H. Shroffe and P. Manimegalai, “Hand gesture recognition based on EMG signals using ANN,” International Journal of Computer Application, vol. 3, no. 2, pp. 31–39, 2013. [7] Y. P. Ivanenko, R. E. Poppele, and F. Lacquaniti, “Five basic muscle activation patterns account for muscle activity during human locomotion,” The Journal of Physiology, vol. 556, no. 1, pp. 267–282, 2004. 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 Journal of Healthcare Engineering Hindawi Publishing Corporation

Loading next page...
 
/lp/hindawi-publishing-corporation/semg-signal-acquisition-strategy-towards-hand-fes-control-sC5BwHaOyO

References

References for this paper are not available at this time. We will be adding them shortly, thank you for your patience.

Publisher
Hindawi Publishing Corporation
Copyright
Copyright © 2018 Cinthya Lourdes Toledo-Peral 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
2040-2295
eISSN
2040-2309
DOI
10.1155/2018/2350834
Publisher site
See Article on Publisher Site

Abstract

Hindawi Journal of Healthcare Engineering Volume 2018, Article ID 2350834, 11 pages https://doi.org/10.1155/2018/2350834 Research Article 1 1 Cinthya Lourdes Toledo-Peral, Josefina Gutiérrez-Martínez , 1 2 Jorge Airy Mercado-Gutiérrez, Ana Isabel Martín-Vignon-Whaley, 3 3 Arturo Vera-Hernández , and Lorenzo Leija-Salas División de Investigación en Ingeniería Médica, Instituto Nacional de Rehabilitación “Luis Guillermo Ibarra Ibarra”, Calz. México- Xochimilco No. 289, Col. Arenal de Guadalupe, Tlalpan, 14389 Ciudad de México, Mexico Facultad de Ingeniería, Universidad La Salle, Benjamín Franklin 45, Col. Condesa, Cuauhtémoc, 06140 Ciudad de México, Mexico LAREMUS, Sección Bioelectrónica, Departamento de Ingeniería Eléctrica, Centro de Investigación y de Estudios Avanzados del Instituto Politécnico Nacional, Av. Instituto Politécnico Nacional 2508, Col. San Pedro Zacatenco, Gustavo A. Madero, 07360 Ciudad de México, Mexico Correspondence should be addressed to Josefina Gutiérrez-Martínez; josefina_gutierrez@hotmail.com Received 11 August 2017; Revised 1 December 2017; Accepted 27 December 2017; Published 14 March 2018 Academic Editor: Kunal Mitra Copyright © 2018 Cinthya Lourdes Toledo-Peral 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. Due to damage of the nervous system, patients experience impediments in their daily life: severe fatigue, tremor or impaired hand dexterity, hemiparesis, or hemiplegia. Surface electromyography (sEMG) signal analysis is used to identify motion; however, standardization of electrode placement and classification of sEMG patterns are major challenges. This paper describes a technique used to acquire sEMG signals for five hand motion patterns from six able-bodied subjects using an array of recording and stimulation electrodes placed on the forearm and its effects over functional electrical stimulation (FES) and volitional sEMG combinations, in order to eventually control a sEMG-driven FES neuroprosthesis for upper limb rehabilitation. A two-part protocol was performed. First, personalized templates to place eight sEMG bipolar channels were designed; with these data, a universal template, called forearm electrode set (FELT), was built. Second, volitional and evoked movements were recorded during FES application. 95% classification accuracy was achieved using two sessions per movement. With the FELT, it was possible to perform FES and sEMG recordings simultaneously. Also, it was possible to extract the volitional and evoked sEMG from the raw signal, which is highly important for closed-loop FES control. muscle activation, and their ability to span region is 1. Introduction curtailed [3]. Neurological disabilities are caused by damage of the ner- Biomedical signals, such as surface electromyography vous system (which includes the brain and spinal cord); this (sEMG), play a significant role in the measurement of the damage results in the loss of capacity to move and manipu- electrical muscle contraction. Plus, its analysis is one of late things, especially if fine movements are required [1]. The the standard procedures used to identify muscle actions effects of many neurological conditions can vary greatly in normal and pathologic conditions. sEMG signals can be from person to person, as well as from time to time for used for various applications, which include identifying neu- the same person. People with neurological conditions, such romuscular diseases, controlling signals for orthotic or pros- as a stroke, may present hand motor impairment and deficit thetic devices [4], anticipating movements of the muscles [5], controlling machines or robots, or detecting hand gestures to in motor execution, severe fatigue and/or weakness, impaired hand dexterity, tremors, spasticity, abnormal muscle syn- improve the quality of life [6]. ergies, and deficit in motor planning and motor learning sEMG patterns during movements exhibit a great deal of [2]. Stroke survivors may have great difficulty to modulate intersubject, intermuscle, and context-dependent variability. 2 Journal of Healthcare Engineering Understanding the sEMG interactions in hand movements is Skin cleaning a challenge [7]. Several researches have been directed to determine the extent to which each muscle participates in Electrodes template each synchronous and time-varying synergies for an individ- for sEMG location ualized human hand motor pattern [8] or to predict the sEMG patterns associated with static hand postures [9]. Recording and stimulation These studies show the importance of considering different electrodes placement intensities and durations of sEMG bursts, temporal patterns, strength of the muscle contraction [10], and muscle synergy Signal acquisition as a framework for sEMG patterns of hand postures. sEMG patterns are used for neuromuscular biofeedback OpenViBE [11], robot-aided [12] training, and neurorehabilitation, as Virtual platform well as to control devices such as neuroprosthesis based on Rest—10 s functional electrical stimulation (FES), to mimic a neuro- Hand open/power grasp/fine pinch/ muscular function for both upper and lower extremities pronation/supination —10 s [13], or to enhance hand motor recovery when physical ther- apy alone is ineffective in stroke patients [14] or with spinal 10 repetitions Functional cord injury [15]. electrical Several techniques have been employed for addressing stimulation Raw sEMG signal human hand movement patterns from sEMG signal. Tech- niques, such as an adaptive neuro-fuzzy inference system Signal processing integrated with a real-time learning scheme and time- frequency features, have been used to identify hand motion Figure 1: Electrode placement using a personalized template to find commands suitable for hand prosthesis control [16]. Ordinal sEMG signal for acquisition task and stimulation location. After pattern analysis is used to describe corrections of sEMG cleaning the skin and placing the electrodes, the isometric recordings during hand open and hand close states. The contraction (hand open, power grasp, fine pinch, pronation, and results suggest that the mutual information analysis has supination) was performed by the subject during 10 seconds, with potential in identifying different hand movements [17]. Usu- 10 seconds for rest. The task was repeated 10 times. A session included a task for each movement. ally, wavelet transformations and artificial neural network classifiers are used for hand movement analysis [10]. The Hilbert-Huang transform is another technique used to control of a neuroprosthesis to aid in motor neurorehabilita- detect, measure, filter, and decompose sEMG signals in tion of patients suffering from a stroke aftermath. order to identify patterns in time, frequency, or space or The presented technique is based on an array of recording and stimulation electrodes on the forearm, used the combination of flexion/extension arm movements. How- ever, the sEMG patterns can present abnormal muscle syn- to acquire sEMG signals from five hand motion patterns ergies and be indistinguishable [18]. This fact could make from six able-bodied subjects, and the effects of this tech- the classification in some stroke patients more difficult; for nique over functional electrical stimulation (FES) and voli- example, a solution proposed in [3] is to use voice recogni- tional sEMG combinations. tion as an auxiliary in a sEMG-driven actuated glove for clinical therapy purposes. 2. Methodology Recognizing sEMG signals with the aim of controlling assisting devices is not only concerned about feature extrac- 2.1. Identification of sEMG Locations. The first step was to tion and classification of signals but the acquisition site is also find the best electrode positioning for sEMG recording. of major importance. This position was found at the belly of the muscle, on the M-wave is an electrophysiological response evoked by upper part of the forearm, which is formed by the following electrostimulation detected in standard sEMG. It has been muscles: brachioradialis, palmaris longus, flexor carpi radia- studied widely in order to verify the functionality of the lis, flexor carpi ulnaris, extensor carpi radialis longus, and stimulation site measurement over the target muscle, extensor carpi ulnaris. Stimulation is performed at the ends which closely relates to muscle fiber recruitment. This of the same muscles. electrophysiologically driven approach is expected to lead In order to make sure that the electrodes were placed on to the identification of selective electrode configurations the same positions for the different trials for each subject, a of an array for functional movements [19]. However, find- personalized template was made. This template was created ing the best electrode configuration for sEMG recording to as follows: for bipolar channel placement, eight spots, where get the right sequence for movement activation still repre- the electrodes would be placed, were allocated and marked sents a challenge. on a piece of acetate paper. Then, the unique physical charac- This paper is related to the acquisition and analysis of teristics of the individual and the positions of five stimulation sEMG signals for active movements and to obtaining usable bipolar electrodes were marked on the same paper. Once the hand patterns with simultaneous placing of recording and places were allocated and the personalized template was stimulation electrodes on the forearm, for the eventual designed, sEMG acquisition was carried out. Journal of Healthcare Engineering 3 Clock stimulator Acquisition client Clock stimulator Clock stimulator Stimulation multiplexer Stream synchronization In|Out|Set Generic stream writer Signal display In|Out|Set In|Out|Set Display cue image Display cue image In|Out|Set In|Out|Set CSV file writer In|Out|Set Generic stream writer In|Out|Set (a) (b) Figure 2: OpenViBE flow diagram used to acquire raw sEMG signal (a); image cue synchronization control (b). This algorithm completes a movement task. Table 1: Stimulation electrode positions for each of the five target Raw sEMG signal movements. Target movement Electrode position Butterworth filter order 2.59-61 Hz Finger and wrist flexors. Flexor carpi Power grasp radialis, flexor carpi ulnaris, flexor digitorum superficialis. Application of Ulnar nerve. Flexor pollicis longus, flexor Daubechies Lumbrical grip digitorum superficialis. wavelet level-8 Finger and wrist extensors. Hand open Extensor carpi radialis. Extensor digitorum. sEMG signal–baseline Baseline Pronation Pronator teres. Supination Supinator. Application of Haar wavelet The subject’s skin was cleaned using an alcohol swab in level-8 order to reduce impedance and have a better coupling for the skin-electrode interface. Afterwards, the template was sEMG envelope placed on the subject’s forearm and marked; these were the spots where the electrodes should be placed. Figure 1 shows this procedure. The electrodes were kept in contact with the skin with a tubular mesh; this also reduced artifacts due to Figure 3: sEMG signal processing algorithm. The signal was filtered cable movements. for 60 Hz, baseline was subtracted through DWT, and the envelope Electrodes were connected to an open-source platform signal that selected the active pattern was obtained. called OpenViBE. This acquired the sEMG signal through a compatible open hardware acquisition device (OpenBCI) 2.2. sEMG Signals Acquisition. Six able-bodied subjects were which was connected to a designer space, where an algorithm included for the acquisition of sEMG signals, their age was designed for trial tasks (Figure 2). ranged from 21 to 33 years old, three males and three OpenViBE configuration was 24 for gain, 250 Hz for females. The subject was sitting in a comfortable position sampling rate, and eight channels for sEMG. The subject was asked to perform the movement shown in a cue image with his/her right arm supinated and leaning on the table. The subject was asked to perform an isometric contraction while it was on the screen. The task started with a rest for five movements: hand open, power grasp, fine pinch, of 10 seconds, and it continued with a ten-second isometric pronation, and supination. While contraction was active, contraction of hand open, power grasp, fine pinch, prona- the forearm muscles that participate in the motion were tion, or supination, depending on the trial. Cue images were shown alternatively until the subject completed ten palpated and located. 4 Journal of Healthcare Engineering Skin cleaning Functional Recording and stimulation electrical electrodes placement using FELT stimulation Signal acquisition OpenViBE OpenViBE OpenViBE Virtual platform Virtual platform Virtual platform Movement Contraction Rest Movement Contraction Rest Movement Contraction Rest Hand open 1 s 3 s Hand open 1 s 3 s Hand open 1 s 3 s Power grasp 1 s 3 s Power grasp 1 s 3 s Power grasp 1 s 3 s Fine pinch 1 s 3 s Fine pinch 1 s 3 s Fine pinch 1 s 3 s Pronation 1.8 s 2.2 s Pronation 1.8 s 2.2 s Pronation 1.8 s 2.2 s Supination 1.8 s 2.2 s Supination 1.8 s 2.2 s Supination 1.8 s 2.2 s × × × 5 repetitions 5 repetitions 5 repetitions Volitional sEMG signal Raw sEMG signal Evoked sEMG signal by FES + evoked by FES Signal processing Figure 4: sEMG signal acquisition for tasks (hand open, power grasp, fine pinch, pronation, and supination) with FES stimulation. An isometric contraction was performed by the subject for each part of the trial. The motion was repeated 5 times per part. A session included 5 repetitions of volitional contraction, followed by 5 repetitions of sEMG evoked by FES, and finally, 5 repetitions of volitional contraction plus the evoked sEMG by the FES stimulation. repetitions of the motion. A session was considered com- pleted when two movement tasks were finished (Figure 1). Table 2: Stimulation parameters for each subject and target All subjects completed two sessions for each of the men- movement. tioned motions. The tasks of sEMG recordings were saved Pulse Pulse Pulse as .csv files that included the information of eight chan- On/Off Subject Movement amplitude width frequency nels and a time vector. time (s) (mA ) (μ s) (Hz) From all the personalized positions, which were based on common regions found for each subject, a universal template PG 10 300 30 1/3 that kept the array for recording and stimulation electrodes LG 10 300 30 1/3 was designed. It was called forearm electrode set (FELT). 1 HO 10 500 30 1/3 SU 10 500 50 1.8/2.2 2.3. Preprocessing, Selection, and Feature Extraction. Each PR 10 500 30 1.8/2.2 sEMG record was imported into MATLAB® environment PG 10 300 30 1/3 for processing. From the .csv files, information of eight chan- LG 12 300 30 1/3 nels and a time vector was extracted. As seen in Figure 3, the signal was cleaned from line interference at 60 Hz by using a 2 HO 10 300 30 1/3 Butterworth filter, order 2, with a 59 to 61 Hz bandwidth. SU 10 300 50 1.8/2.2 After acquisition, data were conditioned using discrete PR 10 300 30 1.8/2.2 wavelet transforms (DWT). An eight-level decomposition PG 10 300 30 1/3 using mother wavelet Daubechies-4 was applied, and the LG 8 400 30 1/3 reconstructed signal was subtracted in order to eliminate 3 HO 12 300 30 1/3 baseline drift [20]; this was equivalent to filter a 0.7 Hz signal. Then, the DWT was applied, again, to an eight-level SU 10 500 50 1.8/2.2 decomposition, but this time a mother wavelet Haar was used PR 10 500 30 1.8/2.2 in order to find the envelope of the signal, which was PG: power grasp; HO: hand open; SU: forearm supination; PR: forearm obtained from its reconstruction. This envelope was used pronation; LG: lumbrical grip, applied through the RehaStim 2 electrical to find the parts of the sEMG signal that represented a stimulator. movement, in this case open hand or power grasp; then it was converted to a logic signal (Figure 3). Journal of Healthcare Engineering 5 ×10 Raw sEMG signal Processed sEMG signal 2.01 70 1.99 1.98 1.97 1.96 1.95 1.94 1.93 1.92 1.91 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 4 4 × 10 ×10 (a) (b) Figure 5: Subject 1, open hand/rest. Comparison of sEMG signal before and after processing using DWT. (a) Raw sEMG signal containing baseline drift and 60 Hz noise. (b) Processed sEMG signal drift-free and visible active and rest patterns. 500 500 0 0 ‒500 ‒500 0 20 40 60 80 100 120 140 160 180 200 0 20 40 60 80 100 120 140 160 180 200 CH1 CH2 (a) 0 0 ‒100 ‒100 ‒200 ‒200 0 20 40 60 80 100 120 140 160 180 200 0 20 40 60 80 100 120 140 160 180 200 CH1 CH2 (b) Figure 6: For subject 1, (a) open hand and (b) power grasp, sEMG processed and envelope signal obtained for active pattern selection. Example for channels 1 and 2 of 8. In order to find the characteristic features of the five target movements, the following parameters were calculated: SD = 〠 x − x , n − 1 mean absolute value—MAV (1), wave length—WL (2), zero i=1 crossing—ZC (3), standard deviation—SD (4), integral of absolute value—IAV (5), variance—V (6), and slope sign IAV = 〠 x , change—SSC (7). i=1 MAV = 〠 x , 1 i 2 V= 〠 x − x , 6 i=1 n − 1 i=1 WL = 〠 x − x , 2 1, x > x , x > x , i i−1 i i+1 i i−1 i=1 SSC = 7 1, x < x , x > x , i i+1 i i−1 1, x <0, x >0, i+1 i n−1 0, else ZC = 〠 1, x >0, x <0, 3 i+1 i i=1 From these parameters, a subset was selected for clas- 0, sification based on separability between movements and 6 Journal of Healthcare Engineering Mean absolute value Wavelength Standard deviation 60 350 70 50 300 60 40 250 50 30 200 40 20 150 30 10 100 20 0 50 10 −10 0 0 12345678 12345678 12345678 Channel Channel Channel Variance Integral of abs. value 4000 400 −1000 −2000 −100 12345678 12345678 Channel Channel Hand open Power grasp (a) ×10 Mean absolute value Wavelength Standard deviation 50 5 70 40 4 30 3 20 2 10 1 0 0 0 12345678 12345678 12345678 Channel Channel Channel ×10 Integral of abs. value Variance 5000 3.5 2.5 1.5 0.5 −1000 0 12345678 12345678 Channel Channel Hand open Power grasp (b) Figure 7: Analysis of window length for (a) 20 ms and (b) 3 s for all features (MAV, WL, SD, IAV, and V) and 8 channels, using data from the 6 subjects. Journal of Healthcare Engineering 7 classification accuracy. The new set of parameters were Table 3: Analysis of the combinations of selected channels and features with best performance during training, for each window used for classification. length. 2.4. Classification. For the classification of sEMG signals, Window Classifier Channels Features feature and window length analysis were performed for the length (s) accuracy (%) eight channels. The sEMG envelope signal was used for selec- 0.02 1–3 WL, SD 80.69 tion of active patterns at the processing stage. From this ~10 s 0.05 1–3 WL, SD 88.23 of sEMG activity, windows of 20 ms, 50 ms, 100 ms, 300 ms, 0.10 1–3 WL, SD, V 91.56 and 500 ms and 1 s and 3 s length, with a 25% overlap, were 0.30 1–3, 7-8 MAV, WL, SD, V 93.86 used to calculate the seven features described in (1), (2), (3), (4), (5), (6), and (7). 0.50 1–3, 7 MAV, WL, SD 95.83 A linear discriminant analysis (LDA) was executed for 1.00 1–4, 7-8 MAV, WL, SD, V, IAV 94.68 sets of two movements following the process described 3.00 1–3, 7-8 MAV, WL, SD, IAV 95.14 ahead, in this case for hand open and power grasp: The bold rows correspond to classification accuracies above 95%. (1) For each subject and each analysis window value, the the FELT was placed accordingly. For each target motion, seven features were extracted for the eight channels; there was a pair of self-adhesive stimulation electrodes for hand open task and power grasp task. (Axxelgard, USA) placed within the FELT, positioned as (2) The resulting 56 features obtained from each window presented in Table 1. were considered as a single trial for each movement. The subject was asked to perform an isometric contrac- tion for five movements: hand open, power grasp, fine pinch, (3) All available trials from the first session (one task pronation, and supination, but this time the trial consisted of per movement) of all subjects were concatenated three parts (Figure 4): movement-wise and randomized afterwards. (4) Label classes for each trial were set as 1 for hand open (1) Five isometric contractions of the selected move- and 2 for power grasp. ment, each lasting one second with three seconds rest (except pronation and supination: 1.8 active to 2.2 (5) For each window length value, the analysis was second rest) performed ten times. (2) Five FES stimulations of the selected movement, each (6) All trials were divided in 70% for a training set and lasting one second with three seconds rest 30% for a testing set. (3) Five isometric contractions during FES stimulations (7) A LDA classifier was trained with the training set. of the selected movement, each lasting one second (8) The trials on the testing set were classified with the with three seconds rest LDA classifier, and its classification accuracy was The algorithm in Figure 4 was performed once for each calculated as the ratio of correctly classified trials movement and subject. The stimulation parameters changed versus the total number of trials. for each movement according to Table 2. All subjects’ data from the first session (combinations of The new records were analyzed for processing the sEMG features, channels and window lengths) that obtained a clas- data because these signals included evoked and/or volitional sEMG as well as the FES stimulus. In order to extract the sification accuracy higher than 90% were chosen as the subset of features used to train the final LDA classifier. Data from sEMG evoked/volitional sEMG from the stimulus artifact, a comb-type filter was applied to eliminate the 30 or 50 Hz sig- the second session, which consisted of hand open and power grasp for each subject, was processed in the same way and nal of the stimulus, by means of a Butterworth filter, order was used to test the LDA classifier. two, with a 29 to 31 Hz or 49 to 51 Hz bandwidths, accord- ingly. All data processing is designed and performed in MATLAB environment. The parameters calculated for these 2.5. sEMG Recording and FES Application: Acquisition and signals are MAV (1) and root mean square (RMS) (8) to Processing. For the trials of sEMG signal acquisition during compare sEMG of evoked and volitional and evoked signals. FES application, the acquisition was performed using the OpenViBE platform and OpenBCI device with the same con- figuration mentioned above. For FES application, a RehaStim RMS = 〠 x 8 2 electrical stimulator (Hasomed Gmbh, Germany) was used i=1 and programmed in an interface developed in Simulink®/ MATLAB Environment. 3. Results and Discussion Three able-bodied subjects out of the six that performed the previous trials without FES, age range from 22 to 34 A personalized template was designed for each subject. years old, two males and one female, were included for These templates were used to successfully locate muscle sEMG acquisition. Their skin was cleaned with alcohol and sites and place electrodes for the second trial, with the 8 Journal of Healthcare Engineering 400 400 200 200 0 0 ‒200 ‒200 0 20 40 60 80 100 120 140 160 180 200 0 20 40 60 80 100 120 140 160 180 200 CH1 CH2 (a) 0 0 ‒500 ‒500 0 20 40 60 80 100 120 140 160 180 200 0 20 40 60 80 100 120 140 160 180 200 CH1 CH2 (b) Figure 8: Subject 1 using FELT: (a) channels 1 and 2 for open hand and (b) channels 1 and 2 for power grasp. BL/60 Hz ‒500 ‒1000 ‒1500 0 50 100 150 Time (s) sEMG (a) Spectrogram 1.5 ‒20 ‒40 ‒60 0.5 ‒80 ‒100 020 40 60 80 100 120 Frequency (Hz) (b) Figure 9: (a) Power grasp sEMG signal recorded from trial (algorithm Figure 4), channel 1. Baseline has been eliminated using algorithm of Figure 2. (b) Spectogram of sEMG signal, where activity in the 30 Hz band for the 2nd and 3rd sets of motions and their harmonics can be observed. advantage of a tenfold reduction in location time, approx- positions to acquire eight sEMG channels and enough imately. Then, the FELT was designed as a universal array place for five bipolar stimulation channels. It is important from all the individual templates. to mention that since the forearm is a small area, it was The main purpose of the FELT was to simplify recording difficult to find the right allocation for all the electrodes and stimulation electrode placing, for a future FES-based (stimulation electrodes are 5×5 cm and recording electrodes are 1 cm in diameter) and still have useful signals that could neuroprosthesis clinical application for stroke aftermath rehabilitation at upper limb and hand. There are not stan- be processed and classified. dardized designs for sEMG recording and FES application. Due to this critical disposition, the electrode locations The sEMG signals acquired for open hand and power grasp from the personalized templates were assessed through the were used to evaluate the right position of the recording elec- sEMG signals obtained by means of signal processing and classification of movements. trodes at the FELT. One of the objectives of this work was to allocate all A baseline drift-free signal was obtained from the raw electrodes keeping the balance between having available sEMG signal during the preprocessing stage (Figure 5). All Amplitude (mV) Time (min) Power/frequency (dB/Hz) Journal of Healthcare Engineering 9 Volitional sEMG Frequency spectrum Evoked sEMG without FES Frequency spectrum 600 30 800 70 400 600 25 60 ‒200 ‒200 30 ‒400 ‒400 ‒600 ‒600 10 ‒800 ‒800 0 ‒1000 0 30 35 40 45 50 55 0 20 40 60 80 100 120 140 70 75 80 85 90 95 0 20 40 60 80 100120140 Time (s) Frequency (Hz) Time (s) Frequency (Hz) ‒20 ‒40 ‒40 ‒80 ‒60 ‒120 16 20 14 ‒100 80 ‒80 60 10 ‒60 10 40 6 ‒40 20 5 2 ‒20 Frequency (Hz) 0 Time (s) Frequency (Hz) Time (s) (a) (b) Evoked and volitional Frequency spectrum sEMG without FES 200 30 ‒200 20 ‒400 ‒600 10 ‒800 5 ‒1000 0 20 40 60 80 100120 140 110 115 120 125 130 135 140 Frequency (Hz) Time (s) ‒20 ‒40 ‒60 80 15 0 Time (s) Frequency (Hz) (c) Figure 10: Power grasp, subject 1, channel 1, sEMG signals of the 3 parts of the trial. (a) Set of 5 isometric contractions of the selected movement, each lasting 1 second with 3 seconds rest. (b) 5 FES stimulations of the selected movement, each lasting 1 second with 3 seconds rest. (c) 5 isometric contractions during FES stimulations of the selected movement, each lasting 1 second with 3 seconds rest. Figure 6 shows an example of two of the eight sEMG sessions from the six subjects were put through this process- ing. sEMG signal in Figure 5(a) has a large baseline, while channels processed and the envelope signal obtained, which Figure 5(b) shows a cleaner sEMG signal despite original shows the active sEMG sections selected. These correspond baseline drifting; also, the differences between each contrac- to open hand and power grasp movements. tion repetition are clearer. From the analysis of the combinations of features, chan- The preprocessing analysis and processing method nel, and window length for all subjects, it was found that only showed that no matter the 60 Hz noise and drifting base- 5 features (MAV, WL, SD, IAV, and V) yield enough infor- line, the signal could be isolated for feature extraction and mation for classification, above 90% accuracy (Figure 7). In classification. It is important to mention that if the acquisi- Figure 7(b), it can be observed that when the length of the tion signal was less contaminated, this process could be fas- window was larger, for features like MAV or SD, it was easier ter and closer to real time for control applications, which to find a clear separation of the value of the parameters. Even emphasizes the need to design and build a specialized the smaller windows, i.e., 20 ms, (Figure 7(a)) performed with acquisition stage in order to start with the best version of an accuracy of 80.69%. Then, it is important to find a com- a raw sEMG signal (which can also consider a configura- promise between window length and classifier performance. tion that allows the simultaneous application of FES, for From this analysis, using 9 out of 10 repetitions of each volitional sEMG extraction). movement per session and considering session 1 for training Amplitude (mV) Amplitude Amplitude (mV) Amplitude (mV) Amplitude Amplitude 10 Journal of Healthcare Engineering sEMG signal evoked and FES application Volitional sEMG and FES application 500 500 0 0 ‒500 ‒500 ‒1000 ‒1000 ‒1500 ‒1500 72 74 76 78 80 82 84 86 88 90 92 110 115 120 125 130 135 140 Time (s) Time (s) sEMG evoked by FES Volitional sEMG and evoked by FES 800 800 400 400 0 0 ‒200 ‒200 ‒400 ‒400 ‒600 ‒600 ‒800 ‒800 ‒1000 ‒1000 72 74 76 78 80 82 84 86 88 90 92 110 115 120 125 130 135 140 Time (s) Time (s) (a) (b) Figure 11: Comparison of sEMG signals between 2 parts of the trial involving FES application. (a) Raw signal including FES (top) and sEMG signal evoked by FES free of the stimulus (bottom). (b) Raw signal including volitional sEMG and FES (top) and volitional sEMG signal and evoked by FES free of the stimulus (bottom). and session 2 for testing, it was found that MAV, WL, and SD Table 4: RMS and MAV values obtained for 3 able-bodied subjects, comparison between sEMG evoked by FES and the combination of features and a 0.50 seconds’ window length were the best volitional and evoked by FES signals. Values obtained from motion combinations for the classifier to perform with only 4 of power grasp, channel 1. channels (CH1, CH2, CH3, and CH7) at a 95.83% classifica- tion accuracy. The results from all combinations can be seen Volitional sEMG in Table 3. sEMG evoked by FES + sEMG evoked by FES Subject Gender This study and analysis was performed to minimize RMS MAV RMS MAV inputs for the classifier, with the aim of getting a closer (mV) (mV) (mV) (mV) approach to a real-time application. This analysis is a classi- 1 Male 147.5061 105.6109 147.4792 104.6412 fication method for multisubjects, used to generate a 2 Male 159.2150 109.0613 169.7005 126.7346 sEMG-driven control for a FES neuroprosthesis application. 3 Female 306.5072 200.5491 215.4075 138.3950 An example of the sEMG signals obtained for subject 1 using FELT, for channels 1 and 2, is shown in Figure 8. It can be observed that even though the signal was noisier for and five bipolar stimulation channels (larger electrodes, this session, the processing algorithm was still able to find 5×5 cm) in the forearm, which is a small area for so many the active sEMG sections. electrodes (a total of 27). Figure 9 shows the signal resulting from the sEMG Signal processing yielded a very clean signal that pre- (evoked/volitional) and FES stimulus signal acquisition using served sEMG components by using DWT and allowed to dif- ferentiate between movements through feature extraction the FELT. Figure 10 shows each set of repetitions of the 3 parts and classification. of the trial. The frequency spectrum and a 3D spectrogram We found an optimal combination between window are presented. length and number of channels and features, at 0.5 seconds, Figure 11 shows the sets of contractions for sEMG evoked with four channels and three features (MAV, WL, and SD), which allowed a more efficient classification in terms of time by FES and those from a volitional sEMG contribution used in order to compare the effects of both conditions. and channels. The RMS and MAV values for each repetition were The stimulation parameters were selected in order to calculated; Table 4 shows an example of these values. generate a complete movement without subject discomfort; however, range of movement is yet to be evaluated. As for signal processing, knowing the stimulus frequency before- 4. Conclusions hand allows the use of a filtering technique feasible for offline and online application. From Figure 10, it is evident that a The design of a personalized template presented in this paper replicates the sEMG signal between sessions. Also, natural sEMG contraction activates the slow fibers of the the forearm electrode set (FELT) resulted from the need muscle, but in the cases of FES application (Figures 10(b) and 10(c)), the fast twitch fibers have a larger contribution to find the correct place for eight sEMG bipolar channels Amplitude (mV) Amplitude (mV) Amplitude (mV) Amplitude (mV) Journal of Healthcare Engineering 11 [8] M. D. Klein Breteler, K. J. Simura, and M. Flanders, “Timing of to the sEMG record. Additionally, the evoked and voli- muscle activation in a hand movement sequence,” Cerebral tional sEMG with FES were similar; however, it should Cortex, vol. 17, no. 4, pp. 803–815, 2007. be considered that the sample was small and that all sub- [9] A. B. Ajibove and R. F. Weir, “Muscle synergies as a predictive jects were able-bodied. Therefore, a protocol with a bigger framework for the EMG patterns of new hand postures,” Jour- sample is needed and it still remains to be seen if these nal of Neural Engineering, vol. 6, no. 3, article 036004, 2009. results hold for patients. [10] S. M. Mane, R. A. Kambli, F. S. Kazi, and N. M. Singh, “Hand Using the FELT, it was possible to perform sEMG motion recognition from single channel surface EMG using recording and FES simultaneously. Moreover, it was possi- wavelet & artificial neural network,” Procedia Computer Sci- ble to extract the volitional and evoked sEMG from the ence, vol. 49, pp. 58–65, 2015. raw signal, which was accomplished without blanking the [11] O. M. Giggins, U. M. Persson, and B. Caulfield, “Biofeedback signal allowing better control techniques to be implemented. in rehabilitation,” Journal of Neuroengineering and Rehabilita- This is highly important for closed-loop FES control. tion, vol. 10, no. 1, p. 60, 2013. In the evoked/volitional sEMG and FES trials, the FES [12] B. Cesqui, P. Tropea, S. Micera, and H. Krebs, “EMG-based stimulus was successfully eliminated from the recorded sig- pattern recognition approach in post stroke robot-aided reha- nal leaving a usable sEMG signal for FES control and other bilitation: a feasibility study,” Journal of Neuroengineering and applications as orthosis, prosthetics, neuroprosthesis, and Rehabilitation, vol. 10, no. 1, p. 75, 2013. other rehabilitation and assistive devices. [13] D. B. Popović, T. Sinkaer, and M. B. Popović, “Electrical stim- ulation as a means for achieving recovery of function in stroke Conflicts of Interest patients,” NeuroRehabilitation, vol. 25, no. 1, pp. 45–58, 2009. The authors declare that they have no conflicts of interest. [14] F. Quandt and F. C. Hummel, “The influence of functional electrical stimulation on hand motor recovery in stroke patients: a review,” Experimental & Translational Stroke Med- Acknowledgments icine, vol. 6, no. 1, p. 9, 2014. The authors would like to thank the National Council for [15] C. H. Ho, R. J. Triolo, A. L. Elias et al., “Functional electri- cal stimulation and spinal cord injury,” Physical Medicine Science and Technology (Consejo Nacional de Ciencia y and Rehabilitation Clinics of North America, vol. 25, no. 3, Tecnología—CONACyT) for supporting the Project CON- pp. 631–654, 2014. ACYT-SALUD-2016-1-272983 as well as Gabriel Vega [16] M. Khezri and M. Jahed, “A novel approach to recognize hand Martínez, M.Sc., for his contributions. movements via sEMG patterns,” in 2007 29th Annual Interna- tional Conference of the IEEE Engineering in Medicine and References Biology Society, pp. 4907–4910, Lyon, France, August 2007. [1] S. W. Lee, K. M. Wilson, B. A. Lock, and D. G. Kamper, “Sub- [17] G. Ouyang, Z. Ju, and H. Liu, “Mutual information analy- ject-specific myoelectric pattern classification of functional sis with ordinal pattern for EMG based hand motion recogni- hand movements for stroke survivors,” IEEE Transactions on tion,” in ICIRA, Intelligent Robotics and Applications, vol. 7506 Neural Systems and Rehabilitation Engineering, vol. 19, no. 5, of Lecture Notes in Computer Science, pp. 499–506, Springer, pp. 558–566, 2011. Berlin, Heidelberg, 2012. [2] P. Raghavan, “The nature of hand motor impairment after [18] Y. Lan, J. Yao, and J. Dewald, “The impact of shoulder stroke and its treatment,” Current Treatment Options in abduction loading on EMG-based intention detection of hand Cardiovascular Medicine, vol. 9, no. 3, pp. 221–228, 2007. opening and closing after stroke,” in 2011 Annual Interna- tional Conference of the IEEE Engineering in Medicine and [3] D. G. Kamper, “Restoration of hand function in stroke and spinal cord injury,” in Neurorehabilitation Technology,D. Biology Society, pp. 4136–4139, Boston, MA, USA, September Reinkensmeyer and V. Dietz, Eds., pp. 311–324, Springer, Cham, 2nd edition, 2016. [19] C. De Marchis, T. S. Monteiro, C. Simon-Martinez, [4] A. Altamirano, A. Vera, R. Muñoz, L. Leija, and D. Wolf, S. Conforto, and A. Gharabaghi, “Multi-contact functional “Multichannel sEMG signal analysis using Hilbert-Huang electrical stimulation for hand opening: electrophysiologically transform to identify time-frequency features,” in 36th Annual driven identification of the optimal stimulation site,” Journal International Conference of the IEEE Engineering in Medicine of Neuroengineering and Rehabilitation, vol. 13, no. 1, p. 22, and Biology Society, Chicago, IL, USA, August 2014. [5] Y. Hou, J. Zurada, and W. Karwowski, “Prediction of EMG sig- [20] G. Vega-Martinez, C. Alvarado-Serrano, and L. Leija, “ECG nals of trunk muscles in manual lifting using a neural network baseline drift removal using discrete wavelet transform,” in model,” in 2004. Proceedings. 2004 IEEE International Joint 2011 8th International Conference on Electrical Engineering Conference on Neural Networks, pp. 1935–1940, Budapest, Computing Science and Automatic Control (CCE), Merida Hungary, January 2004. City, Mexico, October 2011. [6] E. H. Shroffe and P. Manimegalai, “Hand gesture recognition based on EMG signals using ANN,” International Journal of Computer Application, vol. 3, no. 2, pp. 31–39, 2013. [7] Y. P. Ivanenko, R. E. Poppele, and F. Lacquaniti, “Five basic muscle activation patterns account for muscle activity during human locomotion,” The Journal of Physiology, vol. 556, no. 1, pp. 267–282, 2004. 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

Journal of Healthcare EngineeringHindawi Publishing Corporation

Published: Mar 14, 2018

References