4, DECEMBER 2007


Real-Time Classification of Forearm Electromyographic Signals Corresponding to User-Selected Intentional Movements for Multifunction Prosthesis Control
Kaveh Momen, Sridhar Krishnan, Senior Member, IEEE, and Tom Chau, Senior Member, IEEE
Abstract—Pattern recognition-based multifunction prosthesis control strategies have largely been demonstrated with subsets of typical able-bodied hand movements. These movements are often unnatural to the amputee, necessitating significant user training and do not maximally exploit the potential of residual muscle activity. This paper presents a real-time electromyography (EMG) classifier of user-selected intentional movements rather than an imposed subset of standard movements. EMG signals were recorded from the forearm extensor and flexor muscles of seven able-bodied participants and one congenital amputee. Participants freely selected and labeled their own muscle contractions through a unique training protocol. Signals were parameterized by the natural logarithm of root mean square values, calculated within 0.2 s sliding and non overlapping windows. The feature space was segmented using fuzzy C-means clustering. With only 2 min of training data from each user, the classifier discriminated four different movements with an average accuracy of 92.7% 3.2%. This accuracy could be further increased with additional training data and improved user proficiency that comes with practice. The proposed method may facilitate the development of dynamic upper extremity prosthesis control strategies using arbitrary, user-preferred muscle contractions. Index Terms—Adaptive, classification, clustering, electromyography (EMG), fuzzy c-means, prosthesis, real-time, root mean square (rms).

from multiple sites on the forearm to control a powered prosthesis. Past studies have employed two [4]–[7], three [8]–[10], four [6], [11]–[16], and up to eight [17], [18] recording sites with varying levels of success. To control multiple functions, it is necessary to map EMG signals corresponding to different muscle contractions to a variety of prosthetic functions. This mapping has been demonstrated usually through offline pattern recognition [1], [4], [7], [12], [17]–[19] and in a few instances, with real-time classifiers [5], [6], [11], [20]. For the purposes of this paper, we focus only on the latter. For offline approaches, the interested reader is referred to EMG prosthesis control schemes based on linear classifiers [2], [6], [11], [21], artificial neural networks (ANN) [1], [3]–[5], [8]–[10], [12], [14], [15], fuzzy and neuro-fuzzy classifiers [1], [22], [23], and K-nearest neighbour classifiers [7], [9]. A. Real-Time EMG Classifiers A small number of real-time EMG classifiers have appeared in the literature. Focussing on the transient EMG, Hudgins et al. [4], used simple time domain features and a neural network classifier to achieve 90% accuracy in a four-class problem. The main drawback was that contractions had to be exclusively initiated from rest. This restriction prohibited the user from intuitively switching between classes and impeded the coordination of complex tasks involving multiple degrees-of-freedom [12]. Moving away from the transient signal, Englehart et al. developed a “continuous classifier” using wavelet analysis [6] and subsequently with time domain features [11] to process four channels of steady-state EMG signal. They succeeded in classifying four classes of motion with an average of 0.5% error, and six classes with 2% error. In a subsequent work, Englehart et al. [12] further optimized the continuous classifier. They processed four channels of EMG signal, with the task of discriminating six classes of limb movement. They obtained an impressive 93.25% accuracy using a time-domain feature set, with a multilayer perceptron (MLP) classifier. The classes of motion were constrained to six standard able-bodied movements. In a separate study, Nishikawa et al. [5] proposed a novel online learning method for discriminating six different predetermined hand movements on the basis of two EMG signals harnessed from the palmar and dorsal sides of the wrist. Using a variant of the adaptive heuristic critic (AHC) algorithm,



YOELECTRIC prosthetic hands are controlled using electromyography (EMG) signals. These signals originate from the depolarization and repolarization of the muscle membrane during voluntary contractions and can be measured at the skin surface using either dry or wet-type electrodes. The EMG control signal can be derived from a single site [1]–[3] or

Manuscript received February 23, 2007; revised July 29, 2007; accepted August 22, 2007. This work was supported in part by the Canada Research Chairs program, in part by the REMAD Foundation, in part by Ryerson University, in part by the Natural Sciences and Engineering Research Council of Canada, in part by Bloorview Children’s Hospital Foundation, and in part by the Myoelectric service at Bloorview Kids Rehab. K. Momen is with the Toronto Rehabilitation Institution, Toronto, M5G 2A2 ON, Canada. S. Krishnan is with Ryerson University, Toronto, M5B 2K3 ON, Canada. T. Chau is with the Bloorview Kids Rehab, Toronto, M4G 1R8 ON, Canada (e-mail: Color versions of one or more of the figures in this paper are available online at Digital Object Identifier 10.1109/TNSRE.2007.908376

1534-4320/$25.00 © 2007 IEEE

II. the number and types of movements are not known a priori. participants must adhere to a subset of typical contractions associated with normative hand movements of the able-bodied limb (e. 1) Dependence on thresholds. VOL. A simple amplitude-driven inference rule base. 2) Adherence to subsets of admissible movements. potentially demanding significant user training. as demonstrated in [24]. It has been argued that without the capability to dynamically update over time. However. [12]. as reported in [20]. [17]–[19]. wrist flexion/extension. Thresholds have also been defined in the estimation of contraction onset [1]. but must adhere to a predetermined subset of normative movements. these normative movements may be meaningless and unnatural to the congenital amputee. With time domain features [4].. [6]. [5]. As muscle and skin conditions change. we adopt an unsupervised approach to classifier construction. pronation/supination) [3]. [12]. Hence. who may be more proficient at generating other individual-specific contractions. A very low threshold augments the system’s susceptibility to noise. [11]. [19]. However. 4. the user can then choose the preferred assignment of muscle activity to functional output. Since. for example. More recently. Outstanding Challenges Based on the above appraisal of literature. [25]. there are no preset contraction lengths. a nontrivial real-time task.536 IEEE TRANSACTIONS ON NEURAL SYSTEMS AND REHABILITATION ENGINEERING. 15. PROTOCOL FOR SYSTEM AND USER TRAINING TO USER-SELECTED MOVEMENTS The proposed method is outlined in Fig. 1. which require accurately labeled movements for correct partitioning of the feature space. Protocol for system and user training to user-selected movements. the discrimination rate of many EMG classifiers would drop to about 60% over the course of actual usage [5]. [22]. Overview of the Proposed Method We propose a new approach to map EMG signals to different functions in real-time. where the normalization constants are derived from the amplitude statistics of a sufficiently-sized training set. Ajiboye and Weir [20] used heuristic fuzzy logic to classify up to four forearm movements (i. However. Alternatively. finger flexion/extension. One channel of EMG was assigned to one expected movement. [4] and termination [5]. constructed using fuzzy C-means (FCMs). Obtaining such labeled movements necessitates a strict experimental protocol where users can not freely generate contractions of choice. 3) Lack of real-time adaptation. amputees did not participate in real-time testing and the possibility that they might have been more proficient at generating other “atypical” movements was not explored. NO. the effectiveness of a prescribed threshold may be improved by amplitude normalization. The user freely performs forearm movements of choice while the machine automatically determines the most discernible and repeatable muscle signals. one could circumvent the onset threshold by aptly defining a rest class within a pattern recognition framework.e. amplitude thresholds. some outstanding challenges in real-time EMG classification can be identified. In part. due to perspiration and fatigue. [11]. [25].9% accuracy. Generally for prosthesis control. [11]. during both steady state motions and transitions between movements. In this way. . EMG classifiers have typically required the determination of some threshold. ulnar deviation and finger flexion).g. It is not clear whether or not previously reported classifiers are conducive to real-time adaptation to daily variations in EMG signals. B. thresholds need to be adjusted accordingly.. [19]. [12]. achieved real-time classification rates from 94% to 99%. whereby the participants freely select and label their own movements. [6]. 1 and each block is discussed below. this may be due to the prevalent application of supervised classifiers. A trained prosthetist selected four recording sites for able-bodied participants and three sites for amputees. DECEMBER 2007 Nishikwa et al. Based on this subset of usable signals. or movement sequences for training the system. while an excessively large threshold reduces responsiveness to intentional muscle activity. C. thresholds have been required in the calculation of zero-crossings and amplitude inflexion points [4]. systems tested successfully with able-bodied participants may not be suitable for amputees. the user’s unique abilities are accommodated and user training might be minimized. Unlike previous research [4]–[6]. discriminated six forearm movements with 89. wrist extension and flexion. Fig.

Additionally. There was no visual feedback provided during training. The log-transformed feature space. Acquisition of Self-Selected Contractions As most below-elbow myoelectric prosthesis users control their prosthetic appliances via the forearm flexors and extensors. the electrodes were arranged in a differential configuration over the muscle bulk to harness the highest amplitude signals [26]. using . nonlinearly related to contraction feature extraction was peformed in force [28]. This bandpass range encompasses the useful EMG bandwidth previously reported [26]. The electrode locations were photographed for future reference.: REAL-TIME CLASSIFICATION OF FOREARM ELECTROMYOGRAPHIC SIGNALS 537 Fig.8 GHz. Note that movements did not have to be punctuated with rest states nor were any specific motions or movement sequences imposed. we invoked a nonlinear transformation. The real-time throughout the protocol. The raw signals were amplified (GRASS Telefactor 15A54) 5000 times. 1) Feature Extraction: The root mean square (rms) value of each channel was calculated to create a 2-D feature vector. continous stream of data was collected in this manner for each channel. the raw EMG signals were recorded from these two sites. 2. PCI-6014. Rather. Movements were generated at the particant’s own pace without any further prompting. When prompted by the computer. wires running along the table surface to the acquisition equipment were also secured via tape.1. B. pregelled Ag/AgCl GRASS F-E8SD disposable electrodes. It has been argued that the response time of the control system should not introduce a perceivable delay. in this case the natural logarithm.MOMEN et al. With this general prescription in mind. At each recording site. [27]. a 60-Hz notch filter was used to eliminate line noise. Fig. To establish a well-defined and repeatable resting position for each individual. Additionally. to spread the concentrated data points while condensing the highly scattered points. The necessary software for data acquisition was written in MATLAB 7. 3(a). as illustrated in Fig. 1 GB RAM) equipped with a National Instruments. the feature is computationally simple and by the monotonicity of the logarithmic transformation. 2. generally regarded to be roughly 300 ms [6]. 3(b). Example of electrode placement on the forearm flexors of the amputee participant. Extensor versus flexor feature space. Wires emanating from the electrodes were secured to the participant’s forearm via tape. participants were instructed to attend to their forearm movements. at any desired contraction level and without reference to a corresponding prosthetic function. (b) Features space based on natural log of rms values. System Training Training of the system involved the partitioning of the feature space to represent different classes of separable motions. The bandpass filter mentioned above along with the securing of wires to the participant and to the table mitigated contamination due to motion artifacts. [11]. A 2 min. bandpass filtered in hardware with a passband of 30–1000 Hz and then sampled at 2 kHz. demonstrates a more uniform scattering of points compared to the untransformed rms features of an able-bodied participant shown in Fig. A.2. Likewise. 3. we calculated the rms features of each channel within a 200 ms window in real-time. (a) Feature space based on rms values. the participants produced as many different forearm movements as they could naturally and repeatedly create. Additionally. The signals were acquired via a personal computer (P4. Each point is calculated from 200 ms of data. each participant was asked to sit on a chair and to rest his or her arm in a comfortable position. shown in Fig. The low cutoff frequency of the filter was chosen to avoid motion artifact [26]. 16 bit data acquisition board. Participants were not specifically instructed to avoid forearm motion.

(b) Casi rest. However. 2. including the XieBeni index [30] and silhouettes [31]. this would mean reiterating data collection. C. Exploration and Validation After system training. is the number of clusters and is the center of the th cluster. were initially used to determine the target number of clusters . NO. but was generally interpreted by users as the provision of correct classifications within a tolerable delay. (d) Mano estirada adajo. was fixed at 1. Therefore. the system would be retrained. the participants were given one minute to become accustomed at generating the labeled movements. Several different cluster validity algorithms. In the next section. 4. Once a movement was classified as belonging to a certain cluster. . this choice of cluster number is validated by the user. If the user judged the system to be unresponsive to his or her movements after 2 min of exploration. the different clusters were represented as vertical bars. was then used to segment the feature space into where the number of movements was specified by the user.5]. First. the membership of a feature . 2) Clustering: The fuzzy C-means clustering algorithm [29] regions. 15. (a) Rest. the user was given 2 min to explore the relationship between the discovered clusters and forearm movements. is the fuzziness index. After all the movements were named. Each vector was labeled according to the cluster with the highest membership. the corresponding vertical bar would be highlighted. (a)–(f) Movements chosen by the participant.8 in this study. we elected to use the participant-specified number of movements as the target cluster number. (h) Cluster labels superimposed on the clustered feature space. Based on preliminary empirical investigations. For this phase of the protocol. is the center of the th cluster. 1. (e) Puino arriba. An example of six different movements and their corresponding labeled vertical bars are illustrated in Fig. the system classified new movements into one of the learned clusters. 4. which corresponded to their forearm movements. feature vector. The midpoint. ferred choice for many applications. For visual feedback during the exploration stage. the participants generated each of their selected movements in turn and observed which vertical bar was highlighted. VOL. was estimated by [29] vector (2) (1) where is the number of points (feature vectors) in the training is the 2-D sample. The bars were labeled in any language or with any words which would help participants to recall the movements.5. Cluster Labeling and User Training Once satisfied with the system’s classification of movements. Pal and Bezdek [32] showed the best choice for for a large class of problems lies in the in. (c) Supinated punio. system training and exploration until the user deemed the system responsive. is the degree of membership of observation in the cluster . the participants were asked to label the vertical bars. (g) Corresponding bar graph labels selected by the participant. . 4. The fuzzy C-means procedure minimized the generalized sum-of-squares criterion function features were extracted from the corresponding EMG signals in 200 ms intervals. As shown in Fig. Responsiveness was judged subjectively. Cluster labeling.538 IEEE TRANSACTIONS ON NEURAL SYSTEMS AND REHABILITATION ENGINEERING. To facilitate this exploration. DECEMBER 2007 Fig. and is the Euclidean norm measuring the similarity between an empirical data point and a cluster center [29]. Iterative optimization of the objective function (1). has often been the preterval [1. (f) Mano estirada right. preliminary experiments indicated that these methods yielded cluster numbers that were too often incongruent with the user’s perception of the number of achievable movements. This labeling step was necessary for subsequent system testing. D. the where as before is the fuzziness index. Subsequently. was achieved through and the cluster censuccessive updates of the membership tres as prescribed by [29].

In the present study. To estimate the real-time classification error. A congenital amputee without previous experience with myoelectric control was recruited from the general population. ments. Each row reports the accuracy of one movement.e. We recruited seven ablebodied adults and one below-elbow adult congenital amputee. the number. This condition helped to filter out abrupt spurious movements. For the purposes of system evaluation. The left column is the user-assigned labels. the accuracies of the corresponding discernible subset of movements are much higher. Canada). In these instances.. it can be seen that the able-bodied participant created 6 different 4. and sified.7% movements.: REAL-TIME CLASSIFICATION OF FOREARM ELECTROMYOGRAPHIC SIGNALS 539 TABLE I AVERAGE CLASSIFICATION ACCURACIES Subject 2 repeated the test due to misunderstanding of the protocol. a movement by the participant would be considered intentional if it generated five consecutive feature vectors with the same class label. The user was given 3 s from the time of the cue to generate the corresponding movement. . The amputee participant. This suggests that users were reasonably good at estimating the number of distinct movements they created. . created four movements.5% data collection and 1 min of formal user training. after only 2 min of the system with 87. for the movement was given The performance index by TABLE II PERFORMANCE RESULTS FROM PARTICIPANT 8 (THE AMPUTEE)-TRIAL #2 IV. The computer randomly selected one of the movements and highlighted the corresponding vertical bar in yellow. Within the random sequence of cues. We refer to this subset of movements as the discernible subset. The individual movement accuracies for the congenital amputee and an able-bodied participant are exemplified in Tables II and III. The accuracy for each participant was simply calculated by averaging the performance indices over all move. the lack of prior training in our selected participant might present a nontrivial challenge to the automatic separation of EMG signals. is the number of repetitions of each movement. for example. III. we note that while the raw accuracies are low. (3) is the number of times movement was correctly claswhere is the total number of different movements. discernible by the system with 96. who had never used a powered prosthetic hand. respectively. EMPIRICAL EVALUATION The protocol was approved by the Research Ethics Board of Bloorview Kids Rehab (Toronto. the user was repeatedly cued by the computer to perform the labeled movements in random sequence. naming and nature of movements differed among participants. i. In this example. based on his or her own perception of the movement. i. Note that the quantity of movements discernible with greater than 80% accuracy (fourth column) is either slightly less than or equal to the number specified by the user (second column). 5. Results from his second trial are presented. ON. each movement was requested 10 times. All participants provided informed written consent. Subject 8 (amputee) was asked to participate in the study twice to investigate the possible improvement from the first to the second session. RESULTS Table I summarizes the average classification accuracies and the number of discernible movements from all the participants in this study. [33]).MOMEN et al. and 8 grossly overestimated the number of movements. Therefore.. Participants 1. Since participants trained the system based on their self-selected intentional movements.e.7% accuracy. Able-bodied participants consisted of a convenience sample of adults aged 20–30 years with no upper limb deficiencies. recognizable by 13% accuracy. It was particularly desirable to choose an amputee who was not a powered prosthesis user since he or she would most likely be unfamiliar with forearm muscle control. by three or more. It is generally recognized that reliable EMG control necessitates user training (see.

NO. The algorithm suggested that he could perform only four repeatable self-selected movements with more than 92% 4. which usually only open and close. it is conceivable that the proposed algorithm might be rendered adaptive. Unlike previous studies [12]. Classification of Multiple Movements From the second and third columns of Table I. Focus on Natural Movements The training did not impose any restriction on the participant’s choice of movements. These user errors would diminish the accuracy of the system. Incidentally. For example. it was expected that they could reliably recreate the movements as required. Subsequently. Prosthesis Control The proposed algorithm could potentially be used to control current powered prosthetic hands. there was no need to train the system with only the transient or steady-state portions of the signal. Participant #5 later revealed that a subset of the nine movements produced during the initial data collection did not feel natural. This result emphasizes the importance of accomodating movements natural to an amputee (see movement labels in Table II). Previous studies argued that a participant’s EMG corresponding to a specific movement changes over time [5]. that there is indeed improved classification due to dynamic updating. this remains to be demonstrated and validation of the adaptation process. All the participants in this study encountered the system for the first time and trained the system for only a mere 2 min. For example. natural movements. Future studies may investigate the possibility of employing adaptive clustering [34] to address this limitation through the automatic merging of multiple rest classes. 15. While this retraining extends the overall classifier training time. it is somewhat comforting that only a maximum of one retraining was required in the present experiments.540 IEEE TRANSACTIONS ON NEURAL SYSTEMS AND REHABILITATION ENGINEERING. and 7 judged the system to be nonresponsive during the exploration and validation stage. C. VOL. resting one’s forearm with the hand closed is interpreted by the system as a different state than resting with the hand open. he failed to reproduce the same type of movements later in the testing session. The system appeared to be sensitive to variations in the resting state. Although Participant #5 perceived nine different movements during the initial data collection. Therefore. CONCLUSION We have described a new approach to real-time classification of forearm muscle signals corresponding to self-selected intentional movements. our method admits V. as in conventional Otto Bock electrodes. tight fist and soft fist movements expressed by Participant #6. DECEMBER 2007 TABLE III PERFORMANCE RESULTS FROM PARTICIPANT 6 gests that the system may be trainable with self-selected intentional movements from individuals with unique musculature and aytpical soft tissue configurations. expressed nervousness in responding to the computer prompt. sometimes users reported that they performed the wrong motion. The additional movement classes can be mapped to give the prosthesis enhanced functionality. he could not remember how to generate those movements. VI. DISCUSSION A. 4. would be a nontrivial issue. Preliminary tests suggest that it is possible to decrease the 200-ms feature extractor window down to 40 ms without compromising classification performance. A few participants however. Since participants selected and labeled their own movements. 6. VII. during the testing session.3% accuracy. This may have been the result of the user generating movements which he or she could not easily replicate or the fuzzy clustering algorithm settling into a suboptimal local minimum.2 s windows of uniform classification) for classifying a movement was imposed for testing purposes only. This discrepancy in interpretations inflated within cluster variability and accounted for errors relating to the classification of “Rest” signals. it can be seen that all participants except Participant #5 could produce between four and six movements that were discernible by machine with an average accuracy of 75% or better. B. which may be very different from those of typical able-bodied individuals.. five consecutive The current one second delay (0. The surprisingly high accuracy for the lone amputee participant in this study (Participant #8-Trial 2) sug- . Indeed. Such artifacts can be mitigated by embedding both the amplifier and the filters inside the electrode at the skin surface. both motions represent the resting state. Both signal components were captured by the cluster representation. Using well-established dynamic fuzzy clustering principles [34]. the amputee participant expressed extreme satisfaction with the system responsiveness. The important point here is that multimovement discrimination was facilitated with all participants and that the system was able to recommend the most discernible movements to the user. might control the rate of opening or closing the powered prosthetic hand. From the participant’s point of view. When asked by the computer to produce a movement during the test session. Unlike previous work. It is expected that the accuracy of the system would increase further as the user becomes more proficient at generating individual movements. these participants retrained the system a second time.e. LIMITATIONS As the system uses wet-type electrodes and performs amplification distally. Further studies with larger sample sizes are required to validate these observations. Participants 5. the system learned from the participant’s self-selected. i. However. similar to functionality available in conventional myoelectric prostheses. the system is susceptible to motion artifact.

84.-P. Eng. J. Conf. and P. L. Du. Moorleghem. vol.” in Proc. vol. Jul.. Inbar.” in Proc. The method exploits a low dimensional feature space comprised of the natural log of rms signal values from two sites and deploys fuzzy C-means clustering to partition the feature space. pp. Rousseeuw. Hudgins. West Sino Exposition Proc. 2003. vol. 5th ed. Karlik.” Med.. Perth. Vuskovic and S. Circuits Syst. F. 1995. ON. He is currently working toward the Ph. pp. Moradi. 4. China. Mitsukura. B. Canada. 290–293. [2] S.. D. Boston. 309–319. vol. Englehart. Rehabil. 1985. “Continuous multifunction myoelectric control using pattern recognition.” Amer.Sc. Neural Syst. Soc. no. Zhang. and B. Ma. vol. Eng. IEEE Eng. . 10. 370–379. “Classification of finger activation for use in a robotic prosthesis arm. Hussein and M. Parker. 140–141. M. no. Liu. E. [9] H. May 2002. Yang. Sep. 1510–1519.” IEEE Trans. Eng. no. [31] P.MOMEN et al.” in Proc. Commun. Conf.” Electron. 25th Annu. Lundborg. [10] H. Medicine Biol.-H. 2. P.” IEEE Trans. “Classification of motor commands using a modified self-organising feature map. 82–94. Chan. 3. pp. Rehabil. vol. 23. K.-H. H. 1981.” Med. 11. Hargrove. 848–854.. Biomed. “A gaussian mixture model based classification scheme for myoelectric control of powered upper limb prostheses. R. “Classification of hand direction using multi-channel electromyography by neural network. B. F. 2003. Nov. Nishikawa. [24] L. degree in biomedical engineering at University of Toronto. 2003. 2003. 21. B. “A heuristic fuzzy logic approach to EMG pattern recognition for multifunctional prosthesis control. Yu. May 2003. Chan. Biomed. Y. 2539–2544. vol. vol. N. [28] J. vol. Weir. [13] M. pp. Mar. no. Neural Inf. and A. 10.Sc. Conf. [33] J. pp.. F. Int. and P. Lam. Rocon. Housh.A. L.” J. Eng.” IEEE Trans.” IEEE Trans. 9th Int. 2005. pp. “Feature analysis for the EMG signals based on the class distance. pp. A. 2003. Kaveh Momen received the B. Y.” Physiol. Chan. IEEE Eng. pp. W. 431–438. [14] Y. 2. pp. [29] J. 302–311. Syst. Sepulveda. Scott. Kakazu. Hudgins.” IEEE Trans. Eng. pp. 2000. Peleg. 1. pp.. 8. Pattern Anal. based on minimal user training. Robotics Autom. MD: Williams & Wilkins. Cere. Y.” IEEE Trans. Sep. “A comparison of surface and intramuscular myoelectric signal classification. Conf. Jul. Nov. Englehart.. Biomed. Intell. L. Fuzzy Syst. Pons. pp. Taiwan. MA: Kluwer Academic. Eng. Medicine Rehabil. Soc. “EMG classification for prehensile postures using cascaded architecture of neural networks with self-organizing maps. Deluca. 2. and M.-S. pp. real-time control scheme for multifunction myoelectric control. E... 2001. Phys. 2003. [27] J. 311–317. 1801–1811.” IEEE Eng..” IEEE Trans. Markovitz. Hudgins. L. 1. pp. May 2007. Neural Syst.-W. 2002. Balkenius.-P. and C. R. pt. Eng.. pp.. C. Taipei. Biomech.. 3. Appl. 50. vol. 2. 3. 52. Aug. Taipei. 50. Riso. Alci. 220–225. Parker. 13. F Weir. and L. 405–410. Computational Intell. 71. “Virtual reality training and EMG control of the MANUS hand prosthesis. Hudgins. and Y. 280–291.. 1999. [19] Y. vol. “A robust. vol. no. Cancun. 53–65. H. Pattern Recognition with Fuzzy Objective Function Algorithms. 2005. in 1995. Aug.” IEEE Trans. Bueno.” IEEE Trans. H. pp. 123–129. C. Y. Math. 1987. Process. Canada. and N. 8. vol. and G. 750–754. vol. vol.-S. Mexico. 2. no. vol.. H. 2001. A. [3] B. Kumar. D. B. 2002. “A validity measure for fuzzy clustering. 24. C. “Recognition of EMG signal patterns by neural networks. Neural Networks. “On cluster validity for the fuzzy c-means model. degree in electronics engineering from Azad University. 3. “Wavelet based neuro-fuzzy classification for EMG control. Y. Mach. [32] N. 2003. Stevenson. 23rd Annu. M. Toronto. vol. 20. Robotics Autom.. Wagner. pp. Conf. R. Int. vol. 3. 2002 Int. Nov.. Laurell. ON. 21. no. Bioeng. 3. [4] B. 2003. D. 403–413. no. I. [7] D. H. Englehart and B. Kobe. Biomed. [8] N. and T. El-Bialy. Dynamic Fuzzy Pattern Recognition. Jul. pp.A. E. ElBasiouny. Montelius. “Fuzzy EMG classification for prosthesis control. Eng. Honolulu. vol. pp. Angstenberger. and M. 40. Englehart. 48. Boostani and M. no. 5. Danielsen.” in Proc. Saro. Conf. J. 13.” in Proc. pattern recognition. Sep.. 2003 IEEE Int. E. Joint Conf. pp. pp. Japan. Mar. pp. [22] X. Bezdek. Pah. [16] M. Taiwan. Braiman. 3. Yom-Tov. Oct. no. ACKNOWLEDGMENT The authors would like to thank the research participants who eagerly gave of their time REFERENCES [1] F. Jan. 3. 2005. 2001. the classifier recognized four and five different movements with average accuracies of 92. [11] K. “Evaluation of the forearm EMG signal features for the control of a prosthetic hand. M.. K. Bezdek. Nov. S. Symp. D. 283–287. Disability. Robotics Autom. pp. Holmberg. Wong. Englehart. Yokoi. “On-line learning method for EMG prosthetic hand control. no. Hudgins. [26] C. 1087–1089. vol. Wong.. “Intention detection using a neurofuzzy EMG classifier. and M. Tehran. pp. IEEE 29th Annu. [15] Y. M. no. Chengdu.” in The 7th Australian New Zealand 2001 Intelligent Inf.. Akamatsu. G. Akamatsu.D. Muscles Alive. 54. pp. 2002.. Nov. New York: Plenum. pp. [18] F. 2001. J. 2841–2844. vol. 2005. 6.” in 2002 Int. A. and F. Y. Huang. Santa-Cruz. pp. Zhang. WA. Parker. Eng.7 3. “The use of surface electromyography in biomechanics.. [12] K. “Continuous classification of myoelectric signals for powered prostheses using gaussian mixture models. Huang. HI. A. Kandil. Biol. 305–311. Yazama. “Silhouettes: A graphical aid to the interpretation and validation of cluster analysis. and the M. [21] K.. Y. “Automatic EMG feature evaluation for controlling a prosthetic hand using supervised feature mining method: An intelligent approach. 2003 IEEE Int.. no. Medicine Biol. Commun. respectively. Fukumi. Granat. 1497–1502. [20] A. Taher. pp. “Classification of the myoelectric signal using time-frequency based representations. Reynaerts. vol.: REAL-TIME CLASSIFICATION OF FOREARM ELECTROMYOGRAPHIC SIGNALS 541 any muscle contractions preferred by the user and can classify contractions continuously. Meas.” IEEE Trans. Liu. B. 1992. 860–863. K.2% and 79. 2003 IEEE Int. vol. [34] L. B.-S. DeLuca. [5] D. 6–7. Parker. 3. 847–853. Fukumi. 27. P. “A new strategy for multifunction myoelectric control. no. in 2005. 1255–1261. Sep.” in Proc. 15. [17] R. Eng. A. C. no. 5. Zhang. Med. Basmajian and C. M. Pal and J. Iran. vol. J. 1991. Biomed. Schouenborg. 841–846. 11. “A myoelectric prosthesis controller. Dec. “Classification of prehensile EMG patterns with simplified fuzzy ARTMAP networks. Hudgins.-Dec. no. and L. vol. and stochastic processes. Xu. Baltimore. Phys. [6] K. Conf. and M. Mag.9 16. [30] X. T.. Englehart. O. pp. Toronto. Beni. 95–103. “A fuzzy clustering neural network architecture for multifunction upper-limb prosthesis. pp. Rasmy. and R. Liu. vol. 135–163. vol. Conf. Yang.. [25] A. W. Levin. degree in electrical engineering and computer science from Ryerson University. vol. and N. and N. Sep. His research interests include biomedical signal processing.” Robotica. no. [23] S. Ajiboye and R. J. H. Phys.” in Proc. IJCNN ’02. and A. 3. Eng. Eriksson. Japan. X. B.8%. Computational Appl. 2002. Their Functions Revealed by Electromyography. “Optimal selection of time series coefficients for wrist myoelectric control based on intramuscular recordings. D. Chan and K. 1. 1993. “Linearity and reliability of the integrated EMG versus torque relationship for the forearm flexors and leg extensors. Xie and G. C. M. “A wavelet-based continuous classification scheme for multifunction myoelectric control. Matsumura. Using a maximum membership decision criterion. without the need of a rest period between successive contractions. 1384–1387.” in Proc. Biomed. Y. Mitsukura. vol. Hudgins. and C.” J. Englehart. Levinsson.” Technol. E. 7. Jul. Sebelius. Tokhi. K.-T.” IEEE Trans.. 1997. Rehabil. 2001. V. Huang. pp. Nov.

Toronto. Madras. Krishnan is a recipient of the 2007 Young Engineer Achievement Award from Engineers’ Canada. He holds a Canada Research Chair in Pediatric Rehabilitation Engineering. he joined the staff of Bloorview Kids Rehab. After spending some years in industry. DECEMBER 2007 Sridhar Krishnan received the B. Canada. and currently he is an Associate Professor and Chairman of the Department.D. Calgary. degrees in electrical and computer engineering from the University of Calgary. in July 1999.Sc. He holds the Canada Research Chair position in Biomedical Signal Analysis. where he is currently a scientist in the Bloorview Research Institute.D. Tom Chau (S’92–M’97–SM’03) received the BASc. in 1993. His current research interests revolve around innovative access technologies for children and youth with severe motor impairments who are nonverbal. He is also an Assistant Professor in the Institute of Biomaterials and Biomedical Engineering at the University of Toronto. Dr.542 IEEE TRANSACTIONS ON NEURAL SYSTEMS AND REHABILITATION ENGINEERING. degree in systems design engineering in 1998 from the University of Waterloo. ON. degree in electronics and communication engineering from Anna University. 15.E. AB. respectively. NO. 4. VOL. in 1996 and 1999. where he directs the Clinical Engineering Program. and the M. . degree in engineering science in 1992 and the MASc. Canada. and Ph. India. Ryerson University. He joined the Department of Electrical and Computer Engineering. He received the Ph. degree in electrical and computer engineering in 1994 from the University of Toronto.

Sign up to vote on this title
UsefulNot useful