IEEE TRANSACTIONS ON NEURAL SYSTEMS AND REHABILITATION ENGINEERING, VOL. 19, NO.
5, OCTOBER 2011 465
Guest Editorial Brain Training: Cortical Plasticity and Afferent Feedback in Brain-Machine Interface Systems B RAINMACHINE interfaces (BMI) hold great promise for studying sensorimotor processes in the brain, as well as for restoring independence to persons with profound physical disability. Over the past decade, signicant advances have been made in neural interface technology and signal pro- cessing methods [1]. The resulting increase in yield and quality of neural recordings has allowed more reliable and accurate translation of neural signals into motor commands. Parallel efforts by mechatronics researchers have yielded virtually anthropomorphic prosthetic arms and hands that are nally beginning to rival the capabilities of the human arm [2], [3]. While progress in these areas continues, two new areas have emerged recently that may be essential elements for advancing the quality and complexity of movement that can be controlled by a BMI. First, is the need for tactile and proprioceptive feedback to augment vision. Second, is the need to leverage the brains natural adaptive mechanisms to improve BMI training and operation. Tactile and proprioceptive feedback may be particularly important for tasks involving forceful interactions, such as object grasping and manipulation when vision is inade- quate or distracted by a parallel task (as in drinking coffee while answering e-mails; see cover illustration). However, feedback and learning are not independent processes. Somatosensory feedback is known to play a critical role in motor learning [4]. Consequently, it is reasonable to anticipate that afferent feed- back may be particularly important during the early phases of BMI use. Understanding the relation between afferent feedback and learning and applying these principles to BMI control may be crucial for realizing its full potential. This special section is devoted to research into the develop- ment of somatosensory afferent interfaces for conveying tactile and proprioceptive feedback to users and the role of afferent feedback in facilitating motor learning in the BMI framework. This collection of papers demonstrates several different strate- gies for providing somatosensory feedback through patterned electrical stimulation of central and peripheral targets in the ner- vous system. While it may be unrealistic to expect any of these approaches to yield truly natural and complete restoration of so- matosensory function, even crude somatosensory feedback may be useful given the brains remarkable capacity for learning and plasticity. By understanding how best to train the brain, we can improve prosthetic control. Beyond such immediate effects, it may even further yield therapeutic benets by strengthening spared neural connections in the injured sensorimotor system [5][7]. Digital Object Identier 10.1109/TNSRE.2011.2168989 I. NEURAL SUBSTRATES FOR CREATING A SOMATOSENSORY AFFERENT INTERFACE Two of the studies in this special issue examine the de- tectability of intracortical microstimulation (ICMS) inputs to sensory regions of the cerebral cortex, investigating the effect of varied current and waveform in awake animal subjects. Lee Millers group at Northwestern University [8] studies the somatosensory system of monkeys, while Kevin Ottos group at Purdue [9] uses rat auditory cortex. Evaluating the efcacy of sensory stimulation is challenging, particularly in animal models because of the indirect methods that must be employed to determine the perceptual effects of stimulation. Nevertheless, studies like these will be critical to understand the nature of percepts evoked by activation of an inevitably heterogeneous population of neurons by single electrodes, and how multiple electrodes interact. A third study from the lab of Zelma Kiss at the University of Calgary [10] tested the ability of human subjects to detect and interpret thalamic stimulation delivered through a DBS electrode. Working with humans makes it possible to inquire directly into the nature of the evoked per- cepts. In this case, while the quality of percepts was stable and persisting, they were mostly unnatural, a mixture of tingling, vibration and movement across the skin. The authors reach the sobering conclusion that evoking natural percepts with conventional DBS electrodes will be challenging. However, it is possible that microstimulation with arrays of microelectrodes may prove more effective [11]. It is well established that somatosensation is not a simple pas- sive operation, but is instead an active, exploratory process in- volving a complex interplay with the motor system. Two studies pursued the use of electrical stimulation in the active sensing of real or virtual objects. Work by Jose Carmenas group at the University of California, Berkeley [12] 1 showed that rats could learn to locate virtual objects while whisking, by replacing ac- tual whisker contact with appropriately-timed ICMS delivered to the barrel cortex. These studies and those mentioned above show that animals are able to detect sensory cues conveyed by ICMS, however a more detailed description of the perceptual qualities of these sensory inputs cannot be obtained from ani- mals. Studies with human subjects provide the best opportunity for evaluating the perceptual qualities of afferent interfaces. Ken Horch and colleagues at the University of Utah have been testing strategies for restoring sensory feedback to amputees through 1 It should be noted that as an unintended result of its early submission and acceptance, this paper, which was to have been included in this special issue, was inavertently published in the July issue. 1534-4320/$26.00 2011 IEEE 466 IEEE TRANSACTIONS ON NEURAL SYSTEMS AND REHABILITATION ENGINEERING, VOL. 19, NO. 5, OCTOBER 2011 electrical stimulation of peripheral nerves [13], [14]. In [15], two human participants with transradial amputations used a my- oelectric hand prosthesis to grasp objects of varied size and hardness. Force feedback was conveyed by electrical stimula- tion of cutaneous afferents in the nerve stump and enabled one of the subjects to correctly identify objects as being soft, medium, or hard. This subject also reported sensations of nger motion from electrical stimulation of putative proprioceptive afferents and was able to discriminate object sizes as large or small, but was unable to reliably discriminate intermediate sizes. II. NEURAL CODING OF SOMATOSENSORY INFORMATION Each of these studies required subjects to learn completely arbitrary or potentially unnatural stimulus patterns. The neces- sity for a subject to learn a large number of stimulus patterns is an important potential limitation of these approaches. This approach is unlikely to scale adequately without adding an un- acceptable cognitive burden on subjects. To address this con- cern, several studies have been devoted to determining the nat- ural mapping between limb state and neural state, and to de- vising stimulus methods that might best imitate it. Two studies in awake monkeys addressed the encoding of tactile and pro- prioceptive information and the close interplay between these modalities. Work at Arizona State University in the lab of Steven HelmsTillery showed that many cutaneous neurons in areas 3b and 1 encode elements of both touch and motion [16]. The tactile signals arising during haptic exploration appear to un- dergo dynamic modication by an underlying proprioceptive map, yielding interesting postural dependencies in the response properties of putative tactile afferents. Miller et al. showed that most neurons in area 2, whether with cutaneous or deep muscle receptive elds, encode the direction and speed of hand movement, enabling accurate reconstruction of end point kinematics during movement. They also found that neurons with similar receptive elds tend to cluster together, which may give rise to a more coherent motion percept upon stimulation than would otherwise be the case [8]. Whether the intermingled proprioceptive and tactile representation may ulti- mately prove advantageous or detrimental is an important ques- tion that remains to be answered. Weber et al. (University of Pittsburgh, [8])recorded from pri- mary afferent neurons in the dorsal root ganglia (DRG) and S1 cortex of anesthetized cats during passive movement of the hindlimb. Most DRG neurons were found to encode both posi- tion and velocity information for the limb, consistent with pre- vious observations that both cutaneous and muscle afferent neu- rons convey proprioceptive signals for the limb [17]. These nd- ings may explain the intermingling of tactile and proprioceptive information observed in Millers studies of area 2. Webers lab also examined the response of S1 neurons to limb movement and DRG microstimulation using stimulus patterns that were set to mimic the spatiotemporal pattern of activity recorded in the DRG during movement. The S1 responses to microstimula- tion were fairly similar to the S1 responses evoked during move- ment, suggesting that naturally patterned stimulation applied on a relatively small number of channels may be effective in deliv- ering limb-state information to the brain. Since it is practically impossible to create a truly biomimetic afferent interface, it will be important to determine how de- viations from naturalistic patterns degrade performance. A modeling study of the responses of cutaneous mechanorecep- tive afferents innervating the glabrous skin of the hand offers some encouragement (Sliman Bensmaia, University of Chicago, [18]). Despite the heterogeneity of the responses of different af- ferents, a single canonical model could be used to accurately predict the population response to mechanical stimulation. This suggests that even a reduced set of inputs may be capable of conveying rich sensory information. To this end, the work of Karim Oweiss lab at Michigan State University [19] is devoted to the development of novel stimulation protocols designed to combat the lack of selectivity and the heterogeneity of neural responses evoked by electrical stimulation. Stimulation parame- ters are continuously optimized through a feedback control loop driven by motor system recordings that optimize the activation of large networks of neurons using a relatively small number of electrodes and at or near threshold currents. III. BRAIN-TRAINING WITH A BMI Most of the invasive, spike-based BMI approaches begin by attempting to compute an optimal transformbetween neural dis- charge and desired motor output. In contrast, the work of Eb Fetz and Andy Jackson is predicated on the assumption that recording and decoding biomimetic motor commands or en- coding naturalistic feedback through electrical stimulation is essentially impossible [20]. They review a range of their own and others work, and argue that performance using brain or myoelectric interfaces depends almost entirely on the subjects ability to learn the mapping fromthe efference copy of the motor command to its consequence in the task space, a process that may serve as a proxy for missing somatosensation. A key to the difculty of this task may be in the relative dimensionality of the control signals and the task space. Learning to control a pros- thesis can be quite rapid if control is mapped from the activity of muscles to functionally related motions. However, even non- intuitive maps can be learned with enough practice. The more limited dynamic range and spatial resolution of noninvasive BMI systems has traditionally caused more em- phasis to be placed on learning to modulate cortical rhythmic ac- tivity in ways that may not closely mimic those of natural move- ments. Leo Cohens group at the National Institute of Neurolog- ical Disorders and Stroke has been studying the ability of both healthy subjects and stroke patients to learn to control a hand or- thosis using event related desynchronization (ERD) of the sen- sorimotor rhythm (SMR; 1114 Hz band of EEG or MEG) [21]. In this issue, Soekadar et al. [22] show that by coupling in- creasing levels of ERD with different speeds of hand closure, both subject groups adapted dramatically faster than when they had only binary feedback. The question of whether brain ac- tivity closely coupled with somatosensory feedback may give IEEE TRANSACTIONS ON NEURAL SYSTEMS AND REHABILITATION ENGINEERING, VOL. 19, NO. 5, OCTOBER 2011 467 rise to longer term therapeutic effects in addition to improved BMI performance is of great interest. LEE E. MILLER, Guest Editor Departments of Physiology and Physical Medicine and Rehabilitation, Feinberg School of Medicine, and the Department of Biomedical Engineering Northwestern University Chicago, IL 60611 USA DOUGLAS J. WEBER, Guest Editor Department of Physical Medicine and Rehabilitation and the Department of Bioengineering University of Pittsburgh Pittsburgh, PA 15213 USA REFERENCES [1] A. B. Schwartz, X. T. Cui, D. Weber, and D. W. Moran, Brain-con- trolled interfaces: Movement restoration with neural prosthetics, Neuron, vol. 52, pp. 205220, Oct. 5, 2006. [2] C. Cipriani, M. Controzzi, and M. C. Carrozza, The SmartHand tran- sradial prosthesis, J. Neuroeng. Rehabil., vol. 8, p. 29, 2011. [3] L. Resnik, Research update: VA study to optimize DEKA arm, J. Rehabil. Res. Develop., vol. 47, p. ixx, 2010. [4] T. Sakamoto, K. Arissian, and H. Asanuma, Functional role of the sensory cortex learning motor skills cats, Brain Res., vol. 503, pp. 258264, 1989. [5] N. Birbaumer and L. G. Cohen, Brain-computer interfaces: Commu- nication and restoration of movement paralysis, J. Physiol., vol. 579, pp. 62136, Mar. 15, 2007. [6] W. Wang, J. L. Collinger, M. A. Perez, E. C. Tyler-Kabara, L. G. Cohen, N. Birbaumer, S. W. Brose, A. B. Schwartz, M. L. Boninger, and D. J. Weber, Neural interface technology for rehabilitation: Exploiting and promoting neuroplasticity, Phys. Med. Rehabil. Clin. N Am., vol. 21, pp. 157178, Feb. 2010. [7] T. A. Thrasher, V. Zivanovic, W. McIlroy, and M. R. Popovic, Reha- bilitation of reaching and grasping function severe hemiplegic patients using functional electrical stimulation therapy, Neurorehabil. Neural Repair, vol. 22, pp. 706714, Nov.Dec. 2008. [8] D. J. Weber, B. M. London, J. A. Hokanson, C. A. Ayers, R. R. Torres, B. Zaiimi, and L. E. Miller, Limb-state information encoded by pe- ripheral and central somatosensory neurons: Implications for an af- ferent interface., IEEE Trans. Neural. Syst. Rehabil. Eng., vol. 19, no. 5, pp. 501513, Oct. 2011. [9] A. Koivuniemi and K. Otto, Asymmetric versus symmetric electric pulses for intracortical microstimulation, IEEE Trans. Neural. Syst. Rehabil. Eng., vol. 19, no. 5, pp. 468476, Oct. 2011, 2011. [10] E. Heming, R. Choo, J. Davies, and Z. Kiss, Designing a thalamic so- matosensory neural prosthesis: Consistency and persistence of percepts evoked by electrical stimulation, IEEE Trans. Neural. Syst. Rehabil. Eng., vol. 19, no. 5, pp. 477482, Oct. 2011, 2011. [11] S. Ohara, N. Weiss, and F. A. Lenz, Microstimulation the region of the human thalamic principal somatic sensory nucleus evokes sensations like those of mechanical stimulation and movement, J. Neurophysiol., vol. 91, pp. 736745, Feb. 2004. [12] S. Venkatraman and J. M. Carmena, Active sensing of target location encoded by cortical microstimulation, IEEE Trans. Neural Syst. Re- habil. Eng., vol. 19, no. 3, pp. 317324, Jun. 2011. [13] G. S. Dhillon and K. W. Horch, Direct neural sensory feedback and control of a prosthetic arm, IEEE Trans. Neural. Syst. Rehabil. Eng., vol. 13, no. 4, pp. 468472, Dec. 2005. [14] G. S. Dhillon, T. B. Kruger, J. S. Sandhu, and K. W. Horch, Effects of short-term training on sensory and motor function severed nerves of long-term human amputees, J. Neurophysiol., vol. 93, pp. 26252633, May 2005. [15] K. Horch, S. Meek, T. Taylor, and D. Hutchinson, Object discrimina- tion with an articial hand using electrical stimulation of peripheral tac- tile and proprioceptive pathways with intrafascicular electrodes, IEEE Trans. Neural. Syst. Rehabil. Eng., vol. 19, no. 5, pp. 483489, Oct. 2011. [16] L. Rincon-Gonzalez, J. Warren, D. Meller, and S. Helms-Tillery, Haptic interaction of touch and proprioception: Implications for neuroprosthetics, IEEE Trans. Neural. Syst. Rehabil. Eng., vol. 19, no. 5, pp. 490500, Oct. 2011. [17] R. B. Stein, D. J. Weber, Y. Aoyagi, A. Prochazka, J. B. Wagenaar, S. Shoham, and R. A. Normann, Coding of position by simultaneously recorded sensory neurones the cat dorsal root ganglion, J. Physiol., vol. 560, pp. 883896, Nov. 1, 2004. [18] S.-S. Kim, S. Mihalas, A. Russel, Y. Dong, and S. J. Bensmaia, Does afferent heterogeneity matter conveying tactile feedback through pe- ripheral nerve stimulation?, IEEE Trans. Neural. Syst. Rehabil. Eng., vol. 19, no. 5, pp. 514520, Oct. 2011. [19] J. Liu, H. K. Khalil, and K. G. Oweiss, Neural feedback for instan- taneous spatiotemporal modulation of afferent pathways bi-directional bra machine interfaces, IEEE Trans. Neural. Syst. Rehabil. Eng., vol. 19, no. 5, pp. 521533, Oct. 2011. [20] A. Jackson and E. Fetz, Interfacing with the computational brain, IEEE Trans. Neural. Syst. Rehabil. Eng., vol. 19, no. 5, pp. 534541, Oct. 2011. [21] G. Pfurtscheller and A. Aranibar, Evaluation of event-related desyn- chronization (ERD) preceding and following voluntary self-paced movement, Electroencephalogr. Clin. Neurophysiol., vol. 46, pp. 138146, Feb. 1979. [22] S. R. Soekadar, M. Witkowski, J. Mellinger, A. Ramos, N. Birbaumer, and L. G. Cohen, ERD-based online brain-machine-interface (BMI) the context of neurorehabilitation: Optimizing BMI learning and per- formance, IEEE Trans. Neural. Syst. Rehabil. Eng., vol. 19, no. 5, pp. 542549, Oct. 2011. Lee E. Miller (M08) received the B.A. degree in physics from Goshen College, Goshen, IN, in 1980, and the M.S. degree in biomedical engineering and the Ph.D. degree in physiology from Northwestern University, Evanston, IL, in 1983 and 1989, re- spectively. He completed two years of postdoctoral training in the Department of Medical Physics, Uni- versity of Nijmegen, Nijmegen, The Netherlands. He is currently the Edgar C. Stuntz Distinguished Professor of Neuroscience in the Departments of Physiology, Physical Medicine and Rehabilitation, and Biomedical Engineering at Northwestern University, Chicago, IL. His primary research interests are in the cortical control of muscle activity and limb movement, the representation of limb state by the somatosensory system, and in the development of brainmachine interfaces that attempt to mimic normal physiological systems. Douglas J. Weber (M94) received the B.S. degree in biomedical engineering from the Milwaukee School of Engineering, Milwaukee, WI, in 1994, and the M.S. and Ph.D. degrees in bioengineering from Arizona State University, Tempe, in 2000 and 2001, respectively. He is currently an Assistant Professor in the De- partment of Physical Medicine and Rehabilitation, University of Pittsburgh, Pittsburgh, PA. He is also a faculty member in the Department of Bioengineering and the Center for the Neural Basis of Cognition. Previously, he was a Postdoctoral Fellow and then an Assistant Professor in the Centre for Neuroscience at the University of Alberta. His primary research area is neural engineering, including studies of motor learning and control of walking and reaching with particular emphasis on applications to rehabilita- tion technologies and practice. Specic research interests include functional electrical stimulation, activity-based neuromotor rehabilitation, neural coding, and neural control of prosthetic devices. Active projects in his lab include development of somatosensory neural interfaces (SSNI) to record from or stimulate primary afferent neurons in cats and humans, and brainmachine interface studies with magnetoencephalography and electrocorticography in humans. Dr. Weber has been a member of the Society for Neuroscience since 1995.