A motion-classification strategy based on sEMG-EEG signal combination for upper-limb amputees
- Xiangxin Li†1, 2,
- Oluwarotimi Williams Samuel†1, 2,
- Xu Zhang1, 3,
- Hui Wang1, 2,
- Peng Fang1Email author and
- Guanglin Li1Email author
© The Author(s). 2017
Received: 24 June 2016
Accepted: 14 December 2016
Published: 7 January 2017
Most of the modern motorized prostheses are controlled with the surface electromyography (sEMG) recorded on the residual muscles of amputated limbs. However, the residual muscles are usually limited, especially after above-elbow amputations, which would not provide enough sEMG for the control of prostheses with multiple degrees of freedom. Signal fusion is a possible approach to solve the problem of insufficient control commands, where some non-EMG signals are combined with sEMG signals to provide sufficient information for motion intension decoding. In this study, a motion-classification method that combines sEMG and electroencephalography (EEG) signals were proposed and investigated, in order to improve the control performance of upper-limb prostheses.
Four transhumeral amputees without any form of neurological disease were recruited in the experiments. Five motion classes including hand-open, hand-close, wrist-pronation, wrist-supination, and no-movement were specified. During the motion performances, sEMG and EEG signals were simultaneously acquired from the skin surface and scalp of the amputees, respectively. The two types of signals were independently preprocessed and then combined as a parallel control input. Four time-domain features were extracted and fed into a classifier trained by the Linear Discriminant Analysis (LDA) algorithm for motion recognition. In addition, channel selections were performed by using the Sequential Forward Selection (SFS) algorithm to optimize the performance of the proposed method.
The classification performance achieved by the fusion of sEMG and EEG signals was significantly better than that obtained by single signal source of either sEMG or EEG. An increment of more than 14% in classification accuracy was achieved when using a combination of 32-channel sEMG and 64-channel EEG. Furthermore, based on the SFS algorithm, two optimized electrode arrangements (10-channel sEMG + 10-channel EEG, 10-channel sEMG + 20-channel EEG) were obtained with classification accuracies of 84.2 and 87.0%, respectively, which were about 7.2 and 10% higher than the accuracy by using only 32-channel sEMG input.
This study demonstrated the feasibility of fusing sEMG and EEG signals towards improving motion classification accuracy for above-elbow amputees, which might enhance the control performances of multifunctional myoelectric prostheses in clinical application.
The study was approved by the ethics committee of Institutional Review Board of Shenzhen Institutes of Advanced Technology, and the reference number is SIAT-IRB-150515-H0077.
KeywordsAmputee EEG Hybrid interface Motion classification Multifunctional prosthesis Pattern recognition Rehabilitation sEMG Signal Fusion
Multifunctional prostheses are commonly used by upper-limb amputees to restore their lost motion functions. Surface electromyography (sEMG) is a kind of neural signal that contains motor commands and can be non-invasively extracted on the muscle surface of residual limbs. Due to its relative ease of acquisition and abundant content of neural information, sEMG plays an important role in the control of modern motorized prostheses [1–4] and rehabilitation robotics [4–6]. In actually applications, however, the residual muscles after amputations are usually limited, especially in the case of above-elbow amputations. Thus, there exists a dilemma that the less the residual muscles are available for prosthesis control, the more the joint movements would be expected. As a result, multifunctional myoelectric prostheses for above-elbow amputees are still seldom seen on the market [7, 8].
Electroencephalography (EEG) is another kind of neural signal that contains the information related to mental activities of brain but is independent of amputation conditions . Several efforts have been exploited to apply EEG as a brain-computer interface (BCI) for possible applications: Hochberg et al. applied EEG to control robotic arms to perform hand movements for paralyzed subjects, by decoding EEG signals recorded with microelectrode array implanted in the motor cortex ; Gernot et al. developed a non-invasive BCI system based on the steady-state visual evoked potentials (SSVEPs) to control a prosthetic hand, where a varied classification accuracy between 44 and 88% was obtained on able-bodied subjects . However, EEG is usually limited for clinical use due to either the biocompatibility of implantable electrode and signal instability for long term application in an invasive way or the low information transmission rate, low spatial resolution, and high signal variability in a non-invasive way [1, 12].
Multiple-source signal fusion is a possible solution for the problem of insufficient information in prosthesis control [13, 14], where some non-EMG signals are combined to sEMG signals to realize a more precise extraction of motor commands. A method based on the combination of sEMG and near-infrared (NIR) signals was reported , where hand gestures could be identified by adjoining two or more sEMG/NIR sensors. However, the NIR signals are associated to limb movements and thus still depend on the amputation condition. Another alternative non-EMG signal for prosthesis control may be the human voice, which is irrelevant to the limb movements and independent of amputation levels. Some previous works have demonstrated the feasibility of using speech as an additional input to realize a flexible control of multifunctional myoelectric prostheses . However, the human speech is non-task-related and amputees would feel awkward to pronounce commands during prosthesis operation.
In this work, a non-invasive hybrid method was developed, which coherently combined sEMG and EEG signals as a parallel input to classify the upper-limb motions for above-elbow amputees. The proposed method was additionally optimized by channel selections in order to enhance the classification performance with a decreased number of electrodes.
Demographic information of subjects
Stump lengtha (cm)
Time since amputation (years)
Signal acquisition and processing
Both the sEMG and EEG acquisition systems used in the experiments have been integrated with a pre-filter by the producers [17, 18]. Thus the initial sEMG signals were firstly filtered with a band pass from 10 to 500 Hz and sampled at a rate of 1024 Hz; the EEG signals were filtered with a band pass from 0.05 to 100 Hz and sampled at a rate of 1000 Hz. Thereafter, both the sEMG and EEG data were preprocessed offline by using the Matlab® 2010a. Besides, the EEGLAB toolbox (Free Software Foundation, Inc. 1991)  was applied to process the EEG data, where the mean of baseline for each channel was removed and the EEG epochs of each motion class was extracted based on the onset-endpoint lines. For a motion class, each of its EEG epochs consisted of 5-s EEG recordings plus 120-ms EEG recordings before the onset of the movement, totally having 5120 data points. To further improve the quality of EEG recordings, the artifacts from some interferences such as eye movements and blinks in each EEG data epoch were detected based on an adaptive threshold that was equal to the 1.2 times of the mean absolute value of an epoch data . If the absolute value of a data point of an EEG epoch was greater than the threshold, the amplitude of the data point was considered as artifact and would be replaced by the averaged value of the amplitudes of three nearby channels. In addition, a 50 Hz notch filter was used to remove the power-line noise for both sEMG and EEG recordings.
Channel combination and motion classification
32-ch sEMG, where only 32 sEMG electrodes were used;
64-ch EEG, where only 64 EEG electrodes were used;
32-ch EEG, where the first 32 of 64 EEG electrodes were used, locating on the frontal, temporal, and central lobes of cerebral cortex that are associated with thinking and motion functions [21, 22], as shown in Fig. 2c.
Strategy I (S-I), where 32 sEMG electrodes and 64 EEG electrodes were used, leading to a total of 96 channels of input, i.e. 32-ch sEMG + 64-ch EEG;
Strategy II (S-II), where 32 sEMG electrodes and the first 32 of 64 EEG electrodes were used, leading to a total of 64 channels of input, i.e. 32-ch sEMG + 32-ch EEG.
Optimized Strategy I (oS-I), where 20 of the 64 EEG electrodes were selected by the SFS algorithm according to Eqs. (2) ~ (4), and then combined with the pre-selected 10 optimal sEMG electrodes, achieving a total number of 30 channels that was comparable in channel number to the single-signal method of 32-ch sEMG;
Optimized Strategy II (oS-II), where the pre-selected 10 optimal sEMG electrodes were used as the initial channel set S 0 in Eq. (4), based on which 20 EEG electrodes were selected by using Eqs. (2) and (3), and then combined with the pre-selected 10 optimal sEMG electrodes, achieving a total number of 30 channels;
Optimized Strategy III (oS-III), where the pre-selected 10 optimal sEMG electrodes were used as the initial channel set S 0 in Eq. (4), based on which only 10 EEG electrodes were selected by using Eqs. (2) and (3), and then combined with the pre-selected 10 optimal sEMG electrodes, achieving a total number of 20 channels.
Classification accuracies (%) for different single-signal methods over all the subjects
Mean ± Std.
75.8 ± 7.3
68.9 ± 5.9
78.0 ± 11.4
75.1 ± 9.9
87.2 ± 14.8
77.0 ± 6.8
73.9 ± 8.4
72.0 ± 4.8
71.0 ± 11.6
71.9 ± 11.0
86.7 ± 6.3
75.1 ± 6.5
60.7 ± 5.7
60.7 ± 6.8
56.7 ± 7.6
57.0 ± 10.9
79.3 ± 13.4
62.9 ± 9.2
Classification accuracies (%) for the dual-signal method of S-I
Mean ± Std.
93.2 ± 2.2
92.8 ± 3.8
94.2 ± 3.2
86.5 ± 6.2
Mean ± Std.
92.2 ± 5.0
89.4 ± 3.1
91.0 ± 5.9
89.8 ± 5.4
95.9 ± 4.1
91.7 ± 3.5
Classification accuracies (%) for the dual-signal method of S-II
Mean ± Std.
87.3 ± 2.8
90.1 ± 6.4
89.1 ± 7.2
83.4 ± 7.8
Mean ± Std.
86.4 ± 5.6
83.9 ± 3.5
87.3 ± 7.0
85.8 ± 5.0
94.1 ± 7.7
87.5 ± 3.0
Classification accuracies (%) for the optimized dual-signal method of oS-I
Mean ± Std.
85.5 ± 4.3
86.4 ± 6.5
88.5 ± 5.9
78.0 ± 9.6
Mean ± Std.
84.2 ± 6.4
81.5 ± 4.0
83.0 ± 9.7
82.3 ± 5.4
92.1 ± 8.8
84.6 ± 4.6
Classification accuracies (%) for the optimized dual-signal method of oS-II
Average ± Std.
87.4 ± 3.2
89.4 ± 4.6
88.1 ± 5.8
83.2 ± 7.7
Mean ± Std.
85.6 ± 6.2
83.5 ± 2.5
87.4 ± 4.2
84.7 ± 2.9
93.9 ± 6.3
87.0 ± 2.7
Classification accuracies (%) for the optimized dual-signal method of oS-III
Average ± Std.
83.7 ± 5.0
87.9 ± 6.4
84.7 ± 7.3
80.7 ± 8.7
Mean ± Std.
82.6 ± 6.8
78.6 ± 2.3
84.7 ± 6.8
82.9 ± 2.5
92.4 ± 8.1
84.2 ± 3.0
Combining multiple physiological signals is a possible approach to acquire abundant neural information for prosthesis control, where the motion intentions of above-elbow amputees could be decoded more accurately. To improve the control performances of multifunctional myoelectric prostheses, a hybrid classification scheme using EEG signals as an additional augment to sEMG signals for hand and wrist motion identification is studied in this work.
With the single-signal methods that used sEMG recordings or EEG recordings only for motion classification, the 32-ch sEMG input obtains an average classification accuracy of 77.0%, which is 1.9% higher than the 64-ch EEG input and 14.1% higher than the 32-ch EEG input. Thus it might be concluded that the sEMG signals usually outperform the EEG signals in motion classification since the sEMG signals generated by muscle contraction would have a more direct correlation with motions and a higher signal-to-noise ratio in comparison to EEG signals . However, the sEMG is sometimes limited for some motion classes, e.g. the classification accuracy for HO in this work was only 68.9%, which was far from satisfaction for application. This is also confirmed by the subjects’ feedback that HO was the most difficult to perform in the tests. The possible reason would be that when a transhumeral amputee does a hand motion using his lost arm, his residual upper-limb muscles would generate quite weak sEMG signals since these muscles are not electro-physiologically associated with hand movements [21, 34–36].
On the other side, the EEG signals produced by mental activities are less dependent on, or even independent of, the amputation conditions. Therefore, EEG signals would be an effective auxiliary signal for sEMG to improve the pattern recognition performances. In this study, sEMG and EEG signals were firstly combined in two different designs (S-I and S-II) for motion classification. As shown in Table 3, by the dual-signal method of S-I (32-ch sEMG + 64-ch EEG), an average classification accuracy of 91.7% over all the motion classes was achieved. Especially, the classification accuracy for HO was 89.4%, which was 20.5% higher than that by the single-signal method of 32-ch sEMG method. This result has clearly demonstrated that combining some non-EMG signals, such as EEG, to sEMG signals could significantly improve the classification performance for some forearm motions on transhumeral amputees. Different from S-I, the dual-signal method of S-II (32-ch sEMG + 32-ch EEG) used only the first 32 of total 64 EEG channels located on the cerebral cortex area, which is associated with the thinking and motion functions [20, 21]. Compared with the S-I, by which the classification accuracy was improved by 14.7% (from 77.0 to 91.7%), the S-II could still improve the classification accuracy by 10.5% (from 77.0 to 87.5%), meaning that the first 32 EEG channels contributed the most part of performance improvement.
There is generally a direct proportion between the motion classification performance and channel number, as demonstrated by the results of this work (Figs. 4 and 6) and some previous studies [6, 33]. However, a large number of electrodes are not feasible in actual applications partly due to the users’ comfort, computation complexity, prostheses’ weight and cost, and etc., which would eventually affect the acceptability of prostheses [34, 37]. Thus, a balance between the electrode number and control performance is very essential to realize applicable prosthetic systems. The SFS algorithm is a commonly used channel-selection method [33, 34] and it was adopted in this study to optimize the proposed hybrid strategy. According to the channel analysis shown in Fig. 4, for both sEMG and EEG signals, the classification accuracies increased significantly with the increase of channel number until about 10, but the increase slowed down if more channels were further added, which means that the last set of signal channels may not provide much effective information for motion classification . In this work, three optimized strategies of oS-1, oS-II, and oS-III were applied, which decreased the channel number largely at the expense of a slight decrease of classification accuracy compared with S-I and S-II. Besides, it was observed from Fig. 6 that although oS-I and oS-II had the same channel number (both with 10-ch sEMG + 20-ch EEG), the average classification accuracy of oS-I was 2.4% lower than that of oS-II. A possible reason is that in oS-I, the optimal 10-ch sEMG and 20-ch EEG were selected by using the SFS algorithm from the 32 sEMG and 64 EEG channels, respectively, and therefore they might not be the best matching group for motion classification, which is consistent with the conclusion by Cover . Contrarily, in oS-II, the 20 EEG channels were picked out based on the pre-selected 10 optimal sEMG channels, and thus the matching degree between the sEMG and EEG is much higher. In this work, the optimized method oS-II (10-ch sEMG + 20-ch EEG) could obtain an average classification accuracy of 87.0%, which is very comparable to S-II (32-ch sEMG + 32-ch EEG) of 87.5%, but its channel number was less than half of S-II. By oS-III, where totally only 20 channels (10-ch sEMG + 10-ch EEG) were used, the classification accuracy for each subject was above 80.0% with an overall average of 84.2%, which is still about 9.1 and 7.2% higher than those of the single-signal methods of 64-ch EEG and 32-ch sEMG, respectively. These results demonstrated that the proposed strategies have been properly optimized by the SFS algorithm, and the oS-III might be quite desirable for the classification of hand and wrist motions for above-elbow amputees. If higher classification accuracy is desired, the oS-II could be applied. Additionally, the performances of the proposed strategies from TH4 were lower than those for other subjects. A possible reason might be that TH4 has longer and thicker hairs, which leaded to higher impedance between the scalp and the EEG electrodes, resulting in the noisier EEG recordings in comparison to others. So the quality of EEG recordings would be very important for achieving high classification accuracy.
From Fig. 5 it can be observed that the optimal sEMG channels varied among different subjects, which is due to their different status of arm amputations such as different length of remaining arms, different circumstance of arms, and different amputated durations. However, except for few electrodes located on the deltoid muscle for TH1 and TH2, most of the optimal sEMG channels were located on the muscle belly of biceps brachii and triceps brachii of the four subjects. From Fig. 7 it can be seen that the optimal EEG channels were also different in the four subjects, but most of them were located on the contralateral motor cortex. Moreover, for both TH1 and TH2, the EEG channel of POZ located on the visual cortex area contributed greatly in the classification accuracy increase. This is reasonable because the subjects performed motions by looking at the instruction pictures and thus the visual cortex area were easily activated, and TH1 and TH2 who had shorter residual arms might need activate more extra brain function areas such as visual cortex area to complete the motions.
In this study, the results were obtained based on offline analysis, and the performances of the proposed hybrid methods were only evaluated by the classification accuracy. In the future, the efficiencies of hybrid methods, especially the one with optimized channel sets, would be investigated in real-time environment and assessed by more measures such as motion selection time, motion completion time, and motion completion rate [7, 40]. Note that the 20 channels (10-ch sEMG + 10-ch EEG) were finally selected for motion classification, but they may be still too many for the practical applications of the proposed method. For the real-time applications such as myoelectric prosthetic systems, the number of electrodes should be as few as possible to reduce the complexity of the systems and make them more practical, but reducing the electrode number may decreasing the classification accuracy of movements. Thus there is a tradeoff between channel number and classification performance that should be considered in the practical applications. In the future studies, some approaches such as long term training, concentric bipolar electrode, and different classification algorithms would be used to further reduce the channel number and improve the classification performance. In addition, another limitation of the current study is that the proposed method involved EEG recording, which would not be feasible outside the controlled environment of the lab since it is impossible to wear and carry on a commercially available EEG systems such as the NeuroScan in the practical applications. With rapid development of electronics and microprocessors, it is possible now to develop a miniature, portable, and wireless EEG system for the applications of the proposed method in the prosthetic control such as the microEEG device developed by Omurtag et al. .
In this study, hybrid classification methods based on the combination of EEG and sEMG signals were proposed for the classification of hand and wrist motions for above-elbow amputees. By comparing the dual-signal methods with single-signal methods, a significant improvement in classification accuracy was achieved on all the subjects. By using the SFS algorithm, two optimal channel sets of 10-ch sEMG + 20-ch EEG and 10-ch sEMG + 10-ch EEG were obtained with desirable classification accuracies of 87.0 ± 2.7% and 84.2 ± 3.0%, respectively. The results of this study might be helpful in realizing the control of multifunctional myoelectric prostheses for above-elbow amputees.
Brain computer interfaces
EMG pattern recognition
Linear discriminant analysis
Sequential forward selection
Steady-state visual evoked potentials
The authors would like to thank the members at Neural Engineering Center, Institute of Biomedical and Engineering, Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences for their assistance in the experiments.
This work was supported in part by the National Key Basic Research Development Program of China (2013CB329505), the National Natural Science Foundation of China (61135004, 91420301, 61203209, and 51275101), the Natural Science Foundation for Distinguished Young Scholars of Guangdong Province, China (2014A030306029), the Special Support Program for Eminent Professionals of Guangdong Province, China (2015TQ01C399), the Shenzhen High-level Oversea Talent Program (Shenzhen Peacock Plan) Grant (KQCX2015033117354152), and the Science and Technology Planning Project of Guangdong Province, China (#2014A050503052).
Availability of data and materials
The data supporting the conclusions of this article are included within the article.
XL and OWS were involved in experimental design, data acquisition, analysis, interpretation, and drafting of the manuscript. XZ and HW were involved in the data acquisition and analysis. GL and PF supervised the study and was involved in all the stages of the study including critical revision of the manuscript. All authors read and approved the final version of the manuscript.
The authors declare that they have no competing interests.
Consent for publication
All participants gave informed consent to data collection and permission to publish data, results, and photographs for scientific and educational purposes.
Ethics approval and consent to participate
The ethics committee of Institutional Review Board of Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences approved this study. Written informed consent was obtained from all participants.
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.
- Schultz AE, Kuiken TA. Neural interfaces for control of upper limb prostheses: the state of the art and future possibilities. Pm & R J Inj Funct Rehabil. 2011;3:55–67.Google Scholar
- Lobo-Prat J, Kooren PN, Stienen AHA, Herder JL, Koopman BFJM, Veltink PH. Non-invasive control interfaces for intention detection in active movement-assistive devices. J NeuroEng Rehabil. 2014;11:1–22.View ArticleGoogle Scholar
- Jiang N, Englehart KB, Parker PA. Extracting Simultaneous and proportional neural control information for multiple-DOF prostheses from the surface electromyographic signal. IEEE Trans Biomed Eng. 2009;56:1070–80.View ArticlePubMedGoogle Scholar
- Li X, Chen S, Zhang H, Samuel OW, Wang H, Fang P, Zhang X, Li G. Towards reducing the impacts of unwanted movements on identification of motion intentions. J Electromyogr Kinesiol. 2016;28:90–8.View ArticlePubMedGoogle Scholar
- Farina D, Ning J, Rehbaum H, Holobar A, Graimann B, Dietl H, Aszmann OC. The Extraction of neural information from the surface EMG for the control of upper-limb prostheses: emerging avenues and challenges. IEEE Trans Neural Syst Rehabil Eng. 2014;22:797–809.View ArticlePubMedGoogle Scholar
- Geng Y, Zhou P, Li G. Toward attenuating the impact of arm positions on electromyography pattern-recognition based motion classification in transradial amputees. J NeuroEng Rehabil. 2012;9:74.View ArticlePubMedPubMed CentralGoogle Scholar
- Kuiken TA, Li G, Lock BA, Lipschutz RD, Miller LA, Stubblefield KA, Englehart KB. Targeted Muscle reinnervation for real-time myoelectric control of multifunction artificial arms. JAMA-J Am Med Assoc. 2009;301:619–28.View ArticleGoogle Scholar
- Scheme E, Englehart K. Electromyogram pattern recognition for control of powered upper-limb prostheses: state of the art and challenges for clinical use. J Rehabil Res Dev. 2011;48:643–59.View ArticlePubMedGoogle Scholar
- Fouad MM, Amin KM, El-Bendary N, Hassanien AE. Brain Computer Interface: A Review. In Brain-Computer Interfaces: Current Trends and Applications. Edited by Hassanien EA, Azar TA. Dordrecht: Spinger Cham Heidelberg New York; 2015: 3-30.
- Hochberg LR, Daniel B, Beata J, Masse NY, Simeral JD, Joern V, Sami H, Jie L, Cash SS, Patrick VDS. Reach and grasp by people with tetraplegia using a neurally controlled robotic arm. Nature. 2012;485:372–5.View ArticlePubMedPubMed CentralGoogle Scholar
- Müller-Putz GR, Gert P. Control of an electrical prosthesis with an SSVEP-based BCI. Biomed Eng IEEE Trans. 2008;55:361–4.View ArticleGoogle Scholar
- Farina D, Jensen W, Akay M. Introduction to neural engineering for motor rehabilitation. Hoboken: Wiley & Sons; 2013. p. 327-48.
- Huang H, Zhang F, Hargrove LJ, Dou Z, Rogers DR, Englehart KB. Continuous locomotion-mode identification for prosthetic legs based on neuromuscular-mechanical fusion. IEEE Trans Biomed Eng. 2011;58:2867–75.View ArticlePubMedPubMed CentralGoogle Scholar
- Lalitharatne TD, Teramoto K, Hayashi Y, Kiguchi K. Towards Hybrid EEG-EMG-based control approaches to be used in Bio-robotics applications: current status, challenges and future directions. Paladyn. 2013;4:147–54.Google Scholar
- Herrmann S, Attenberger A, Buchenrieder K. Prostheses control with combined near-infrared and myoelectric signals. Lect Notes Comput Sci. 2011;6928:601–8.View ArticleGoogle Scholar
- Fang P, Geng Y, Wei Z, Zhou P, Tian L, Li G. New control strategies for multifunctional prostheses that combine electromyographic and speech signals. Intell Syst IEEE. 2015;30:1–1.View ArticleGoogle Scholar
- Debener S, Minow F, Emkes R, Gandras K, Vos MD. How about taking a low-cost, small, and wireless EEG for a walk? Psychophysiology. 2012;49:1617–21.View ArticlePubMedGoogle Scholar
- Jiang N, Rehbaum H, Vujaklija I, Graimann B, Farina D. Intuitive, online, simultaneous, and proportional myoelectric control over Two degrees-of-freedom in upper limb amputees. IEEE Trans Neural Syst Rehabil Eng. 2014;22:501–10.View ArticlePubMedGoogle Scholar
- Delorme A, Makeig S. EEGLAB: an open source toolbox for analysis of single-trial EEG dynamics including independent component analysis. J Neurosci Methods. 2004;134:9–21.View ArticlePubMedGoogle Scholar
- Nolan H, Whelan R, Reilly RB. FASTER: Fully automated statistical thresholding for EEG artifact rejection. J Neurosci Methods. 2010;192:152–62.View ArticlePubMedGoogle Scholar
- Diers M, Christmann C, Koeppe C, Ruf M, Flor H. Mirrored, imagined and executed movements differentially activate sensorimotor cortex in amputees with and without phantom limb pain. Pain. 2010;149:296–304.View ArticlePubMedGoogle Scholar
- Raffin E, Mattout J, Reilly KT, Giraux P. Disentangling motor execution from motor imagery with the phantom limb. Brain j neurol. 2012;135:582–95.View ArticleGoogle Scholar
- Li G, Schultz AE, Kuiken TA. Quantifying pattern recognition—based myoelectric control of multifunctional transradial prostheses. IEEE Trans Neural Syst Rehabil Eng. 2010;18:185–92.View ArticlePubMedPubMed CentralGoogle Scholar
- Englehart K, Hudgins B. A robust, real-time control scheme for multifunction myoelectric control. IEEE Trans Biomed Eng. 2003;50:848–54.View ArticlePubMedGoogle Scholar
- Boostani R, Moradi MH. Evaluation of the forearm EMG signal features for the control of a prosthetic hand. Physiol Meas. 2003;24:309–19.View ArticlePubMedGoogle Scholar
- Hudgins B, Parker P, Scott RN. A new strategy for multifunction myoelectric control. IEEE Trans Biomed Eng. 1993;40:82–94.View ArticlePubMedGoogle Scholar
- Li X, Chen S, Zhang H, Zhang X, Li G. Effects of non-training movements on the performance of motion classification in electromyography pattern recognition. In: International conference of the IEEE engineering in medicine & biology society. 2014.
- Phinyomark A, Phukpattaranont P, Limsakul C. Feature reduction and selection for EMG signal classification. Expert Syst Appl. 2012;39:7420–31.View ArticleGoogle Scholar
- Geethanjali P, Mohan YK, Sen J. Time domain feature extraction and classification of EEG data for brain computer interface. 2012. International Conference on Fuzzy Systems and Knowledge Discovery.View ArticleGoogle Scholar
- Greene BR, Faul S, Marnane WP, Lightbody G, Korotchikova I, Boylan GB. A comparison of quantitative EEG features for neonatal seizure detection. Clin Neurophysiol. 2008;119:1248–61.View ArticlePubMedGoogle Scholar
- Hargrove LJ, Li G, Englehart KB, Hudgins BS. Principal components analysis preprocessing for improved classification accuracies in pattern-recognition-based myoelectric control. IEEE Trans Biomed Eng. 2009;56:1407–14.View ArticlePubMedGoogle Scholar
- Pan L, Zhang D, Sheng X, Zhu X. Improving Myoelectric control for amputees through transcranial direct current stimulation. IEEE Trans Biomed Eng. 2015;62:1927–36.View ArticlePubMedGoogle Scholar
- Daley H, Englehart K, Hargrove L, Kuruganti U. High density electromyography data of normally limbed and transradial amputee subjects for multifunction prosthetic control. J Electromyogr Kinesiol. 2012;22:478–84.View ArticlePubMedGoogle Scholar
- Geng Y, Zhang X, Zhang Y-T, Li G. A novel channel selection method for multiple motion classification using high-density electromyography. Biomed Eng Online. 2014;13:1–16.View ArticleGoogle Scholar
- Wang H, Fang P, Tian L, Zheng Y. Towards determining the afferent sites of perception feedback on residual arms of amputees with transcutaneous electrical stimulation. In: International conference of the IEEE engineering in medicine & biology society. 2015.Google Scholar
- Woodhouse A. Phantom limb sensation. Clin Exp Pharmacol Physiol. 2005;32:132–4.View ArticlePubMedGoogle Scholar
- Liu L, Su Z, Wang R, Luo X. Material-aware cloth simulation via constrained geometric deformation. Comput Graph. 2013;37:21–32.View ArticleGoogle Scholar
- Peng HC, Long FH, Ding C. Feature selection based on mutual information: Criteria of max-dependency, max-relevance, and min-redundancy. IEEE Trans Pattern Anal Mach Intell. 2005;27:1226–38.View ArticlePubMedGoogle Scholar
- Cover TM. The best Two independent measurements Are Not the Two best. IEEE Trans Syst Man Cybern. 1974;SMC-4:116–7.View ArticleGoogle Scholar
- Huang H, Zhou P, Li G, Kuiken TA. An analysis of EMG electrode configuration for targeted muscle reinnervation based neural machine interface. IEEE Trans Neural Syst Rehabil Eng. 2008;16:37–45.View ArticlePubMedPubMed CentralGoogle Scholar
- Omurtag A, Baki SGA, Chari G, Cracco RQ, Zehtabchi S, Fenton AA, et al. Technical and clinical analysis of microEEG: a miniature wireless EEG device designed to record high-quality EEG in the emergency department. Int J Emerg Med. 2012;5(1):35.View ArticlePubMedPubMed CentralGoogle Scholar