Review For reprint orders, please contact [email protected]

Neural machine interfaces for controlling multifunctional powered upper-limb prostheses Kengo Ohnishi†, Richard F Weir and Todd A Kuiken

CONTENTS Overview of upper-limb loss Neural machine interfaces for artificial upper-limb control Alternative control stategies Expert commentary Five-year view Key issues References Affiliations

†

Author for correspondence Visiting Professor, Northwestern University, Prosthetic Research Laboratory, Research Associate, Oita University, Department of Welfare Engineering, Faculty of Engineering 700 Dannoharu, Oita, 8701192, Japan Tel.: +81 97 554 7771 Fax: +81 97 554 7507 [email protected]

KEYWORDS: amputation, congenital limb deficiency, electroencephalogram, electromyogram, electroneurogram, multifunctional control, neural interface, upper limb, prosthesis

www.future-drugs.com

This article investigates various neural machine interfaces for voluntary control of externally powered upper-limb prostheses. Epidemiology of upper limb amputation, as well as prescription and follow-up studies of externally powered upper-limb prostheses are discussed. The use of electromyographic interfaces and peripheral nerve interfaces for prosthetic control, as well as brain machine interfaces suitable for prosthetic control, are examined in detail along with available clinical results. In addition, studies on interfaces using muscle acoustic and mechanical properties and the problem of interfacing sensory information to the nervous system are discussed. Expert Rev. Med. Devices 4(1), 43–53 (2007)

The goal for a prosthetic arm is to regain sensorimotor function, to restore the appearance of a limb, and to improve the quality of life of people with upper-limb loss. The human arm is a complex anatomical and physiological structure; unfortunately current prosthetic technology is a poor substitute. The commercial electric prosthetic hands are generally single-degree-of-freedom (opening and closing) devices that are usually implemented with control-utilizing bioelectric signals generated as a by-product of healthy muscle contraction. Prosthetic arms with multi-degree-offreedom (multiple joint motions) control most often use sequential control, switching from one joint motion to another to accomplish diverse tasks. This operational method is slow, cumbersome and tiring. People with recent hand amputations hope that modern hand prostheses can be similar to artificial hands, as depicted in science fiction movies. With this great gap between users’ hope and reality, upper limb prostheses are frequently rejected by the amputees or only occasionally used as a tool. The goal of upper-extremity prosthetics research is the purposeful, subconscious control of movement using a multi-functional artificial arm and/or hand. The challenge is to develop durable, multifunctional, mechanical limbs

10.1586/17434440.4.1.43

with controllers that can be interfaced to the human body so that they can run in real-time with accurate processing algorithms that drive the joints simultaneously. In the near future, using advances in materials, motors, computers and design methodologies, we anticipate the development of improved mechanical devices. A major factor limiting the development of more sophisticated upper-limb prostheses is the difficulty of having sufficient control for the many degrees-of-freedom required to replace a physiological hand or arm. This review presents an overview of the current status and issues involved in neural machine interfacing (NMI) to gain advanced prosthesis control. Overview of upper-limb loss Epidemiology of amputation & congenital limb deficiency

Amputation of the upper limb commonly occurs in healthy, young, adult males who have sustained a work-related injury, frequently to the dominant extremity [1]. Traumatic amputation is by far the most common cause of upper-limb loss. Specific causes of amputation vary from country to country [2] or culture. In the USA, the leading causes of trauma-related amputation are injuries involving machinery, powered tools and appliances, fire arms and

© 2007 Future Drugs Ltd

ISSN 1743-4440

43

Ohnishi, Weir & Kuiken

motor vehicle crashes [3]. Recently, there has been a great deal of attention paid to the relative increase in amputation in war, as soldiers are wearing body armor that allows them to survive blasts, but their unprotected limbs suffer severe injury. A study analyzing hospital data in Maryland for all patients discharged with a diagnosis of trauma-related amputation, upper and lower, over a 15-year period (1979–1993) reported that among the 6069 patients with this diagnosis, 4108 (68%) had upper-extremity amputation injuries [3]. The same study demonstrated that, in 1993, the annual incidence rate relative to the population of the targeted region was broken down as follows: For major upper limb extremity amputation injuries: • 0.04 per 100,000 persons/year for above elbow • 0.08 for below elbow • 0.02 for multiple upper-limb extremity For minor amputation: • 3.19 per 100,000 persons/year for hand • 0.52 for finger Most rates had decreased from the 1979 high. A study of the 10-year period (1990–1999) in Swedish cities reported that a total of 21,085 open hand, wrist or forearm injuries occurred during 2,183,677 person-years [4]. The injuries were classified into three levels: minor, moderate and severe. Swedish incidence rates of severe injuries (involving amputation through the hand and more proximally) were 7.5 per 100,000 person-years. Their rate is higher as a result of degloving, crush, penetrating, major laceration of arteries and other injuries proximal to the hand. This study identified a difference in incidence rate between the sexes with 11.1 for males and 4.0 for females per 100,000 person-years. Their incidence of upper extremity amputation or devascularization injuries that potentially required replantation and revascularization was 1.9 per 100,000 person-years and incidence of amputations at, or proximal to, the wrist was 0.11 per 100,000 personyears, which tallies well with the incidence rate reported by [3]. The majority of injuries were partial hand (including finger) amputation, followed by transradial and then transhumeral amputations. This study demonstrates that, although there are chances of replantation at the hand and digit level, most severe injuries proximal to the wrist result in amputation. Congenital-limb deficiency is the other primary reason for upper limb loss. In the USA, the prevalence of upper-limb deficiency varied slightly by region, rates ranging from 2.8 per 10,000 births in blacks of the Metropolitan Atlanta region between 1983 and 1988, to 6.0 per 10,000 births in whites from Maryland in 1984 [5]. The rates of reported congenital-limb deficiency across different countries fell within 2–7 per 10,000 live births, and there are no trends over time. The rate of upper-limb deficiency was 2- to 3-times greater than that of lower-limb deficiency. Prescription & fitting externally powered upper-limb prosthesis

Generally, a upper limb prosthesis consists of a terminal device, joint mechanism, socket, control interface, and a suspension system. There are three broad categories of upper limb prosthesis:

44

passive (cosmetic), body-powered, and externally powered. The components used in these devices are selected by the patient and the rehabilitation support team, based on the client’s physical status, amputation level, cognition, premorbid lifestyle and acceptable cost [2]. The priority for assembling the components is set by the assessment of control sites and the control methodis set by the prosthetist. Body-powered components are light, durable and have limited sensory feedback through the harnesscable system, but require physical ability to generate force and excursion for operation. As the name suggests, body-power harnesses the body’s own strength to provide the power to drive this form of prosthesis. Usually a harness consisting of a loop worn about the axilla of the contralateral limb is used to anchor a control cable that is then attached via a Bowden Cable (similar to a bicycle cable) to the prosthesis. Biscapular abduction (shrugging ones shoulders) or glenohumeral joint flexion extends the cable and transmits that excursion and force to the terminal device. The other option is to use components that are driven by an external power source, such as electricity, pneumatics or some other source of power external to the body. For all intents and purposes, all externally powered devices available today are electrically powered and use batteries as their power source. Externally powered components are chosen depending upon limb length, limb strength, tissue condition and voluntary controllable electromyographic signal sites. The electromyogram (EMG) is an electrical signal that is generated as a by-product of normal muscle contraction. The amplitude of the signal is broadly proportional to the level of contraction of the muscle, and, as such, it can be used to control a prosthesis. This form of control is termed myoelectric control. Often, myoelectric components are chosen because they can often be used with less harnessing around the shoulder, they provide a better aesthetic result, as well as output force independent of physical ability. Major drawbacks are the lack of tactile and somatosensory information for controlling the device, durability and weight. Also, with the under-actuated forearm and wrist, the human-like shaped hand is difficult to orient for picking up objects because it blocks the user’s vision. Hybrid systems, a combination of body-powered and myoelectric components, are commonly used for high level amputations (amputation at or above the elbow). They allow two joints to be controlled at once – one with body power and one with myoelectric control. They are generally less heavy and less expensive than a completely externally powered system [6]. All surveys of upper-limb amputees indicate that current prosthetic devices are greatly in need of improvement; persons with upper-limb loss at any levels are not satisfied with the prehensile function of their terminal device, especially on tasks for which they would have used their biological hand. Transradial electric-powered users would prefer to have more independent movements in the digits (bending fingers and a thumb that moved out to the side) and less visual attention during task performance [7]. In a follow-up study of prosthetic use among upper limb traumatic amputees, the satisfaction and use rate of

Expert Rev. Med. Devices 4(1), (2007)

Neural machine interfaces for multifunctional powered upper limb prostheses

adult traumatic amputees is lower than that of the young congenital amputees. The study reported that vocational and leisure needs were lacking in the devices used by adult traumatic amputees [8]. As these reports demonstrtated, limb loss in later life is more difficult to adapt to, so the limited performance of the prosthetic hand is not satisfactory. In another study, it was found that partial hand amputees have a higher level of disability than major unilateral upper limb amputees, but less than bilateral upper limb amputees [9]. Partial hand amputees tend to experience greater challenges, possibly due to the lack of components available for this level of amputation. The report also discusses the level of amputation and how it affects disability in their leisure and work tasks. A current prosthetic hand which opens/closes the fingers and thumb lacks dexterity; not only in its mechanism, but even more in interfacing the user’s intention for precise control. Partial hand amputees who have muscles connected to the CNS in the forearm have a major advantage over higher level amputees. Unfortunately, current myoelectric prostheses are still rudimentary and are frequently not used. The idea of using these sites to improve the interface between user and prosthesis is not new. The following is a review of recent research in the area of NMIs for upper-limb prosthetic device control. Neural machine interfaces for artificial upper-limb control

The most intuitive way to gather control information is to tap into the human neural control system. The use of neural bioelectric signals for control of artificial arms has been extensively studied. With voluntary movement, bioelectric signals can be recorded throughout the neural axis, including from muscles, nerves and central neurons. The phenomenon of signals captured from muscle is the electromyogram (EMG). The potential captured from the peripheral nerves and neurons is the electroneurogram (ENG). The electroencephalogram (EEG) is the electrical potential measured on the surface of the skull. The EEG is a by-product of normal brain function. Bioelectric signals can also be recorded from the cortex of the brain. These signals are often termed the electrocorticogram (ECoG) or the intracranial EEG (iEEG), although they are sometimes termed local field potentials (LFPs) primarily in the experimental animal literature. The range of parameters and signal frequencies are different, with each signal requiring electrodes and amplifiers that are designed to capture that specific signal. Adequate signal processing procedure to stably extract information content to correspond to the intended motions of the prosthesis is also required. Finally, the system, as well as the amputee, requires a certain amount of adjustment and training to change their interface. Interface through the muscle

Since muscles are the most distal extension of the motor nervous system, and their anatomical structure and relation to the joints are determined, the myoelectric signal is an accessible control source for prosthesis control. The EMG signals may be picked up with electrodes on the body surface as well as

www.future-drugs.com

intramuscularly. However, dry surface electrodes are currently the only practical way to receive myoelectric signals for prosthesis control as the electrodes must be used for long periods of time every day and, hence, must be benign to skin and tissues. Myoelectric control is the current state of the art of NMI for upper-limb prostheses. Good reviews of myoelectric control are by Parker and Scott [10] and the discussion of myoelectric signals by Basmajian and Deluca [11]. Myoelectric control of prostheses is particularly applicable for fitting transradial amputation levels, where the electrodes are placed over the residual antagonistic flexor and the extensor muscles are used to close and open the prosthetic hand. This enables the user to relate original muscles with the prosthetic hand movements. The most common clinically available myoelectric control uses two sites to operate the joints [12]. The amplitude or the changing rate of the rectified myoelectric signal is used to control the motor action or to switch between joints. If the amplitude of the signal can be controlled voluntarily in a wide range, the speed of the hand can be operated in proportion to the signal level. However, this control strategy is not appropriate for multifunctional or multi-degree-of-freedom devices required by higher level amputees. The more joints you have to control, the more the operation becomes sequential and cumbersome. One method for overcoming this issue is to eliminate additional switching operations by detecting characteristics in the signal for specific movements and using them to drive the prosthesis. Pattern recognition of myoelectric signals is a potential solution for this method. Another alternative method is to use multiple electrodes and implantable electrodes to retrieve independent signal sources. In addition, a new surgical technique is now being used to create additional EMG control site. Recent advances in high-speed digital electronics have enabled algorithms of high computational complexity to be run in real time. Complex time, frequency and time-frequency identification techniques have been used to extract features from myoelectric signals for multifunction prosthesis control. Hudgins and colleagues were the first to report on a patternrecognition based approach that offered real time performance with high accuracy [13]. A formal scheme for acquisition and analysis of multiple EMGs for prosthetic device control consists of feature extraction, dimensionality reduction and pattern recognition [14]. Many investigators have conducted experiments attempting to achieve pattern recognition with high classification accuracy. Just to mention a few of the techniques that have been reported recently: self-organizing maps [15,16], genetic algorithms [17], fuzzy logic classifiers [18,19], wavelet analysis [20], Hidden Markov model [21], Gaussian mixture model [22], and Recurrent Log-Linearized Gaussian Mixture Network [23]. Reviews [24] and [25] are good sources of information of multifunctional control and pattern-recognition strategies. Many of the processing techniques described require intensive computation. As a result of this, there is concern regarding the processing time required for these complex algorithms [26]. In order to gain higher recognition accuracy, many studies set the maximum acceptable delay at 200–300 ms.

45

Ohnishi, Weir & Kuiken

Recent work by Farrell has found that processing delays of greater than 100 ms produce a statistically significant degradation in performance [27]. One of the unanswered questions regarding multiple surface EMG pattern recognition techniques is that they have not yet been used for clinical purposes. To enhance the use of these techniques in clinical applications it may be necessary to evaluate the user’s ability to control these advanced systems. These evaluations might include using virtual environments [28,29], and other measurement devices, such as joint movement detection gloves on the sound side [30], to train and evaluate the system. The main problem from a clinical perspective with surface EMG electrode systems is their lack of robustness to variances caused by donning, fatigue, perspiration and other conditions that cause changes in the electrical characteristic of the signal sites. These difficulties can be overcome by methods such as online learning to re-train the network [31] and multi-channel signal recording methods [32,33]. However, further studies in actual environment testing are required to understand their limit of application. Generally, only two independent (free of cross-talk) surface EMG sites can be obtained on the residual limb, even in the transradial amputee where there are many muscles remaining in the residual limb that perform different functions in the wrist and hand. With intramuscular EMG, the signal from individual muscles can be recorded, making a greater amount of control data available. Thus, the use of implanted myoelectric sensors should enable many more independent EMG sites in the residual limb and offers a means of providing simultaneous control of multiple degrees-of-freedom in a multifunction prosthetic hand. Implantable myoelectric sensors (IMES) are being designed for implantation into the muscles of the forearm [34,35]. IMES utilizes the hermetically sealed capsule and electrodes of BION® [36,37] and will transcutaneously-couple, via a magnetic link, to an external exciter/data telemetry reader. The external exciter/data telemetry reader consists of an antenna coil laminated into a prosthetic interface so that the coil encircles the IMES. Another concept is a surface recorded intramuscular EMG (SRI EMG), which is a completely embedded passive conductor that transmits intramuscular EMG signals to a subcutaneous terminal just beneath the skin’s surface at a distant location [38]. This system takes advantage of the current surface EMG electrode technology in an attempt to obtain more control data. At levels of amputation above the elbow, ‘targeted muscle reinnervation’ [39,40], offers a promising new surgical technique to create additional EMG control sites, using the residual nerves in an amputated limb. With this technique residual peripheral nerves are transferred to expendable regions of the muscle in or near an amputated limb. The nerves reinnervate this ‘target’ muscle and produce additional EMG signal sites that are physiologically appropriate to lost-arm function. The reinnervated muscle essentially acts as a biological amplifier of the residual nerves and, thus, the nerves can control functions in the prosthesis that are directly related to their normal

46

anatomical function. Targeted muscle reinnervation has been successfully applied with one bilateral, high-level subject to control a shoulder disarticulation prosthesis and transhumeral amputees [41,42]. One of the drawbacks of the EMG systems is their limited tactile and somatosensory information for controlling the device. One way to close this feedback loop is by means of conscious sensation of tactile information provided through an afferent neural interface, which will be discussed later. Another approach for reducing the user’s mental burden is by applying a hierarchical artificial reflex structure that automates the control during prehension [43]. By allowing the processor to take control, the mental load experienced by the operator is reduced, especially the visual feedback control. The former method assists the user in gaining control over the prosthesis more aggressively and to have more focus on the device, whereas the latter approach removes the user from the control loop to divert his attention to other joints. Controllers with slippage detection are clinically available [101] and attempts to combine the pattern recognition technique and the automated grasping system for multiple-degree-of-freedom hand control are in laboratory testing [44]. In addition, systems that transduce volitional myoelectric signal to control force [45] and joint stiffness [31] are proposed for improving grip and manipulation with the prosthetic hand. Interface through the peripheral nerve

Peripheral nerves transmit command signals between the CNS, muscles and the body’s sensors. Therefore, recording the neural efferent signals can be used for the motion control of prosthesis. Furthermore, sensory feedback from the sensors on the prosthesis may be provided to the user through stimulation of the afferent nerve within the residual limb. Neuroprostheses [46] are developed to artificially substitute or mimic sensorimotor function and interface the peripheral nervous system or muscles by means of appropriate electrodes. Cochlear implant, an auditory prosthesis, is one of the few used and validated artificial interfaces of the nervous system. The key elements for interfacing with the peripheral nerve are electrodes. Clinical peripheral nerve electrodes can be classified broadly into three groups: intraneural, extraneural and miscellaneous. External electrodes provide simultaneous interface with many axons in the nerve, whereas the intraneural and regenerative electrodes only contact small groups of axons within the nerve fascicle. Intraneural electrodes are those that penetrate the nerve fascicle with a needle or wire, such as longitudinally implanted intrafascicular electrodes (LIFE) [47–49], and Utah multiple electrode arrays [50,51]. The Utah Electrode Array (UAE) has a grid of up to 10 × 10 electrodes and was originally developed for recording and stimulation of cortical nervous tissue. This planar structure was later modified to a 3D electrode arrangement as Utah Slant Electrode Array (USEA), to better suit the peripheral nerve. The experimental results of these electrodes for prosthetic control are discussed in the next section. Extraneural electrodes are those in which the active site is placed on the source of the peripheral nerve; cuff electrodes

Expert Rev. Med. Devices 4(1), (2007)

Neural machine interfaces for multifunctional powered upper limb prostheses

are this type. Just to name few, a quasi-tripolar cuff ENG amplifier-telemeter [52] and tripolar nerve cuff with shape memory alloy armature [53] have been reported. De Luca introduced the concept of a neuralelectric control system for upper limb prostheses with a wire-based nerve cuff electrode in 1979 [54,55]. The first advantage of this design is that it is easier to implant and position on the nerve, and the second advantage is that the stimulating current is limited to the inner space of the electrode. The main disadvantage is its configuration. The mismatch of the mechanical properties of the electrode and nerve can easily damage nerve tissues. Another design involves coaxing the surgically-severed nerve to grow into the multiple holes on the electrode. Examples of such electrodes include regeneration microelectrode array [56], regeneration tube [57], sieve electrode [58] and polyimide regenerative electrodes [59]. Edell [60] reported the first neural signal recording obtained with a silicon substrate regeneration microelectrode array designed for amputated peripheral nerve stumps. Material and fabrication technology has improved the biocompatibility and stableness of the electrode, however, there is a conflict in designing the hole size for elongating the regenerating axons. Enlarging the diameter reduces the damage to the motor nerves, however, selectivity needs to be traded off. In the review [61], different types of peripheral nervous systems are discussed concerning their effectiveness as well as their biological, technological and material science issues. Recently, two groups have tested neural control for prosthetic-limb control applications. Longitudinal intrafascicular electrodes (LIFE) were implanted in the peripheral nerves of eight people with transhumeral amputations. These electrodes were used to elicit tactile or proprioceptive sensation by stimulating the nerve, and were also used to record neural signals with which to control a cursor on the computer screen. The results showed that the elicited sensation in the phantom hand could be systematically controlled through modulation of stimulus frequency and amplitude. In addition, through learning and practice, the subjects were able to improve motor control outflow to nerve fibers that had not been connected to muscle for periods of months or years [62,63]. Another project reports an able-bodied subject testing the application of an ENG electrode (Utah electrode arrays) to open/close the prosthetic hand with force feedback neural stimulation [64,65]. A needle-array electrode was surgically implanted into the median nerve fiber of the subject’s left hand. With training, the closing force and timing of the artificial hand improved. Additionally, the force values were most stably controlled when visual and stimulation feedback was provided. However, mechanical failure of the electrode wire bundle caused the experiment to be discontinued. Electrode development is still an active area of research and ranges from early developmental work to proof-of-concept in clinical trials. With the advances of microfabrication technology, electrode development is now moving from handmade devices to micro–electro–mechanical system (MEMS). New technology provides the advantages of better process control, reproducibility, flexibility in design at microscales and enables

www.future-drugs.com

minimal dimension. For clinical use, these electrodes are required to be chronically reliable and safe. Significant numbers of devices in long-term in vivo tests must be assessed before these systems will gain confidence in clinical applications. We believe for practical upper-limb prosthetic usage of the implanted electrodes, the system must be designed by means of telemetric data transfer, as percutaneous wires have a greater risk of being prone to infection and breakage under daily use. Interface through the brain function

The brain is the source of intention for body movement as well as the processor of sensory information. The ready availability of affordable high performance personal computers and an increasing need from people with severe body paralysis has focussed renewed attention on the idea of creating a direct link between the brain and the outside world for augmented communication and control. The EEG is generated as a by-product of the firing of millions of neurons in the brain. With EEG control, the goal is to extract electrical signals from the brain with sufficient bandwidth and at a favorable signal-to-noise ratio, whilst providing a command signal from the cortex. This area of research specifically emphasizes brain–computer interface (BCI) and brain–machine interface (BMI). EEG signals can be obtained in a noninvasive manner through scalp recording. Single or multiple electrodes are mounted on the head or assembled in head gear. Signal features are extracted from a time- and frequency-domain. With advances in BCI, 2D-cursor control is enabled. The review by Wolpaw and colleagues is a good source for general information of current status [66]. The progress and issues of recent BCI research is discussed in reviews [67] and [68]. When recorded with a noninvasive method, the EEG signal represents a field potential generated by the synchronous activity of a large number of neurons, rather than specific cellular activity. Its resolution is temporally and spatially limited, and the rate at which information, which is the user’s intent, is extractable from the EEG signal is limited. Currently, the maximum information rate is 25 bits/min or 2 words/min [66]. Using invasive methods, ECoG signals can be recorded from neurons within the cortex. The electrodes are designed to capture the action potentials of many individual neurons, especially those that code movement or its intent. Motor cortical neurons can provide reliable estimates of motor intentions. Additonally, it has been demonstrated that microstimulation of the somatic sensory cortex can substitute for skin vibration in a perceptual task. This will be discussed later. Electrode arrays can be placed on the dura encapsulating the brain or on the surface of the motor cortex itself (as opposed to penetrating the cortex). Reviews [69–71] are recommended for overviews of BMI and invasive methods. Owing to the constraints in the data acquisition methods, the application and use of the invasive electrodes are limited. Gaining access to the action potential of individual neurons with invasive methods is challenging since surgery is required to place the electrode arrays and the microelectrode tips that are required in close proximity to the signal source. To obtain a

47

Ohnishi, Weir & Kuiken

successful signal, electrodes must remain stable for long periods, or robust algorithms must be identified to deal with shifting populations of neurons. Moreover, dealing with more complex actions, or the simultaneous control of multiple, independent body parts will probably require more electrodes and more arrays. As with other chronically-implanted electrode arrays, to date, the main drawback is that implanted electrodes have a limited functional life. There remains a problem with the fibrotic foreign body reaction that displaces neurons. This causes recorded signals to become smaller and less discriminable. As a control system, the BCI/BMI system has three stages: data acquisition, feature extraction and translation algorithm. During feature extraction, the digitized signals are subjected to one or more of a variety of feature extraction procedures, such as spatial filtering, voltage amplitude measurements, spectral analyses, spike counting or single-neuron separation. Linear and nonlinear methods are used as a translation algorithm. Furthermore, the algorithm is required to adapt to each user on three levels: algorithm’s adoption to the user’s signal feature, periodic online adjustments to the spontaneous variation and accommodation and engagement to the adaptive capability of the brain [66]. This is described as a research platform design for testing closed-loop control BMI [69]. This idea is important since there are two systems, the user’s brain and a computer, that are trying to mutually adapt. Additionally, since the potential changes in the brain are a substitute for the actual neural signal that encodes movement, the user must learn to relate this arbitrary signal to an intended action. Furthermore, the signal is attention-related, which means that use of the bioelectric signal of the brain can interfere with other activities and control can be degraded by distraction. To date, as a result of the remaining disadvantages, there has not been an attempt to apply this interface to an amputee. However, a similar application of the EEG controller for handgrasp control was examined on a neuroprosthesis user [72] and a tetraplegic patient follwoing high-level spinal cord injury [73,74]. The neuroprosthesis user tested the EEG-based neural command for a neuroprosthesis that controls open/close signals of a hand. A scalp recording of the frontal beta rhythm, which was found to be independent to extremity movement, has been tested for the neuroprosthesis. The result demonstrated accuracy rates over 90% at the end of 20 training sessions over 6 months. However, the controller brought up the question of stability over day-to-day use, acceptable delay in command, discrete level for activation and selectivity for practical use. Another concern is the effect of cortical plasticity following injury. A mechanical-hand orthosis was controlled by the tetraplegic patient with ongoing EEG activity based on a synchronous BCI design and two types of motor imagery. After a number of training sessions with varying types of motor imagery strategies over a period of several months, the motor imagery of foot movement and right hand movement achieved an accuracy of close to 100% for controlling the opening and closing of the hand. By means of mentally-induced 17-Hz oscillations as a simple brain switch and by training for six

48

consecutive days, the patient was capable of performing six open/close operations in 1 min, which was about the same as his EMG control capability. Due to disadvantages of the low information rate, BCI and BMI use is limited to people with severe neural disorders with minimal useful motor function, such as persons with high-level quadriplegia. If there is major progress in an information transfer rate, it may be applicable to high-level quadruple amputees or forequarter amputees in extreme condition. Even in such cases, the controllable prosthetic device is limited to simple functions. Alternative control stategies Interface through acoustic & mechanical measures of muscle

The human body has various control options that can be measured including biomechanical displacements and forces, use of muscle acoustic noise and bioelectric signals. As for alternative control options of electrophysiological measures, myopneumatic (pressure distributions from muscle contraction) [75,76], myoacoustic (mechanomyography, MMG) [77,78] and myokinemetric [79] methods have been revisited for use in prosthetic control utilizing advanced sensor arrays and signal processing technologies. To detect the bulging of the superficial extrinsic tendons from the forearm skin surface, tendon-activated pneumatic (TAP) sensors were fabricated from porous polyurethane. The pressure changes between the socket-shaped sensory array and skin were detected. Three amputees tested the interface over a short time period. It proved to be easy to learn and was effective in producing voluntary flexions of three independent fingers and grasping [75]. The pressure sensing method was extracted by applying spatially distributed array sensors (M-P sensor) and a matrix filter to decode the feature signal for discriminating specific finger-flexion commands [76]. The MMG signals are low-frequency sounds or vibration associated with muscle contraction. Using this signal as an interface has advantages such as nonspecific sensor placement, distal-signal measurement and robustness to changing skin electrical impedance. As in EMG recordings, the tolerance to environmental noise is improved by filtering and pattern recognition techniques [77] or by the use of differential signals [78]. An MMG based system was developed to control a commercially available myoelectric hand with no significant delay [77]. Two amputees tested the system over a short time period and gave a positive response. An identified shortcoming of their MMG system was errors to the low frequency noise caused by the movement of the socket and blindness to the strong contraction and a susceptibility to external noises. The myokinemetric signal is measured as the dimensional change of the muscle beneath the skin. Displacement of a magnet on the skin representing the movement of the muscle is picked up by a Hall-effect sensor in the socket. This signal has potential for use in proportional control of joint angle [79]. As in some neural interfaces, these methods also need a great deal more research and development to overcome vulnerable characteristics in signal acquisition and to gain potential over

Expert Rev. Med. Devices 4(1), (2007)

Neural machine interfaces for multifunctional powered upper limb prostheses

the current surface EMG techniques. Methods that use mechanical output of muscle may provide more distinguishable signals than EMG, since subjects can generally use proprioceptive feedback to adjust their level of effort. However, a major problem is the inability of the system to differentiate between actual control signal from a tendon and external pressure impact and vibration caused by the objects in the environment. In activities of daily living, a person wearing prosthesis will exert forces and moments on the socket that may actuate the sensors and incorrect commands to the drive system. Owing to variations in the physical condition of traumatic amputation, these interfaces should be reserved for alternative or combined usage. Interfacing sensory information to the nervous system

One of the major advantages of the externally powered prosthetic systems is its potential to operate multiple degrees of freedom – if the control is available. The disadvantage is the loss of sensory feedback though the cable in the body-powered system. Therefore, current myoelectric prostheses rely on feedback provided primarily by visual monitoring of the prosthesis and pressure exerted by the socket as the prosthesis is used in tasks. To a lesser degree, mechanical noise and the variations of vibration yield some sensory feedback. There is no tactile or proprioceptive feedback with current devices and improvement in the quality of feedback to the CNS is highly desirable. A relatively easy way of doing this is to create a substitute sensory feedback loop. Pressure, vibration and thermo-feedback to the residual limb [80]; and auditory feedback [81] are reported with some success. However, the unnatural feedback signal requires training to correlate the information and stimuli. Moreover, conscious translation and the sensory tolerance may change when the stimulation is applied chronically. To date, sensory substitution has not seen significant clinical application. Further discussion of sensory substitutes for prosthetic control are found in [83]. Targeted sensory reinnervation holds promise for providing sensory feedback. Essentially the afferents of the amputated residual nerves are allowed to reinnervate an area of denervated skin in or near the residual limb. When this skin is touched, the amputee perceives the sensation in the missing limb. Target sensory reinnervation may, thus, be used as a conduit to allow an amputee to feel what they are touching with a prosthesis as if it were in their missing hand [82]. Prosthetic arm control with sensory feedback may be possible with direct nerve interfacing [84]. Dhillon and Horch were able to produce a sense of touch and proprioception in amputees wearing a prosthesis by stimulating afferent axons with LIFEs. The subjects were capable of sensing graded, discrete sensation of touch or movement referred to their phantom hands [85]. There are many challenges to overcome before clinical application; however, the technique holds much promise. Similarly, the potential exists to produce useful sensory perceptions from stimulation of the pre-motor cortex in the brain. This work is in its early stages as it is difficult to assess sensory perceptions in animal models [86].

www.future-drugs.com

Expert commentary

As described in this review, there are three major methods of recording bioelectric phenomena to interface the human intention for voluntarily controlling an upper limb prosthetic device. In the process of transforming bioelectrical signals to output control commands it should be recognized that the multiple structures of the system are the same. Therefore, synergistic effects should be expected from further interaction with neural engineering/science. The feature extraction and classifier algorithms in myoelectric [24,25] and brain machine control should be beneficial to ENG control. Furthermore, the closed loop research platform and algorithm’s three-level adaptability of BMI [66] should be considered in all controllers. This review focuses mainly on neural interfaces for the voluntary control of multifunctional upper limb prostheses. In addition, this review discusses neural interfaces for sensory feedback. Over the years, different designs and concepts for each method and level have been carried out and have demonstrated successful results in experimental environments. However, true success for a prosthetic system can be measured only through clinical use and follow-up studies. Defining clinical success is very difficult and being able to compare systems is also very difficult. Better multilevel objective testing techniques, or a keystone and relational mapping between the evaluation methodologies among multiple institutions is required. This can be beneficial at multiple levels, including controller and mechanical designs, and sensory feedback. All neural interfaces have the potential to be useable for all levels of amputation; yet, broad testing in practical and chronic conditions, is necessary to determine reliability and optimal subjects. In each case, residual function of muscles and nerves needs to be tested to fit the interface. Traumatic amputations usually complicate matters. Since it is not always clear what functional elements are left, such amputation may accompany brachial plexopathy or other nerve injury that is not apparent because the limb is missing. Therefore, fitting the interface requires further knowledge of the nervous system and bioinstrumentation, and multiple diagnoses may be required to assess the residual limb function and select a suitable interface. One final factor that should be discussed for the neural interface to the upper-limb prosthetic control is whether they are preferable or suitable for users with congenital limb deficiencies. Extended physiological proprioception (EPP) has been effective for this group [87]. However, it is unclear if useable neural pathways exist in congenital limb deficiencies. Research quantifying brain activation in prosthesis control, as in [88], can lead to further understanding of prosthesis control for this group of patients. Five-year view

Currently, surface EMG electrodes are the only clinically available product in neural interfaces. Years of modifications has made the surface EMG system more practical and the number of clinical cases is expected to grow. Targeted reinnervation has been successful for transhumeral and shoulder disarticulation amputees. This technique rewires the nerves to the muscle and uses the muscle as an amplifier to interface the neural signal to

49

Ohnishi, Weir & Kuiken

Key issues • The major cause of upper limb amputation is traumatic injury. Another significant group is people with congenital limb deficiency. • Current prostheses are generally limited to sequential control of each degree of freedom. Control is not intuitive, especially for high level amputees were the need is greatest. Furthermore, sensory feedback is generally limited to vision only. Clearly, improvements in prosthetic devices are required to meet the needs of this important patient population. • Several promising neural interface systems are being investigated, including targeted rreinnervation, direct nerve control and brain-machine interfacing. Transducers must be designed to meet the characteristic of the bioelectric phenomena that will be recorded. • Advanced signal processing systems are needed for any of the neural interface systems to be effective. Pattern recognition techniques that allow multifunctional control share a common approach. Active information exchange is mutually beneficial. • Enhanced sensory feedback is a critical issue for improved control, overall function of prostheses and user acceptance. • Better outcome measures need to be established to understand the clinical results in follow-up studies.

the electrode. Therefore, this method is expandable, if healthy nerve and muscles are presented and can be utilized. Clinical results of transhumeral and shoulder disarticulation amputees should be observed and assessed for further development. Pattern recognition EMG system has made progress in experimental settings and further clinical results are expected. Intramuscular or implanted EMG electrodes are in development and laboratory testing should proceed like the test cases for ENG electrodes. Tests for biocompatibility and longer functional life are essential. ENG and EEG based prosthetic control are not yet at a stage to confirm practical use. However, laboratory experiments using these systems show steady progress. We expect that further clinical testing will confirm the adaptability of amputated nerves for the voluntary ENG control; and further studies on sensory feedback will be conducted. As for EEG-based control, improved information transfer rate should be obtained before conducting clinical trials with amputees. References Papers of special note have been highlighted as: • of interest •• of considerable interest 1

2

3

4

50

Esquenazia A, Meier R. Rehabilitation in limb deficiency. 4. Limb amputation. Arch. Phys. Med. Rehabil. 77, S18–24 (1996) Esquenazia A. Amputation rehabilitation and prosthetic restoration: from surgery to community reintegration. Disabil. Rehabil. 26(14/15), 831–836 (2004). Dillingham TR, Pezzin LE, MacKenzie EJ. Incidence, acute care length of stay, and discharge to rehabilitation of traumatic amputee patients: an epidemiologic study. Arch. Phys. Med. Rehabil. 79, 279–287 (1998). Atroshi I, Rosberg H-E. Epidemiology of amputations and severe injuries of the hand. Hand Clin. 17(3), 343–350 (2001).

5

6

•• 7

8

9

As we improve the ability of a user to control artificial arms, improvement in prosthetic components are greatly needed. A multifunctional hand that stands up to chronic use in a daily environment is required. There are no commercially available shoulder components. Without these devices, improved control systems will not get out of the laboratory. Also new socket designs and fitting techniques that can provide a stable base for advanced arms and new neural interface systems are required, whilst maintaining donning/doffing and wearing comfort. As discussed previously, there will be a greater need for sensory feedback with increasing the number of joints to be controlled and their dexterity. It is essential to form a natural closed-loop control of the prosthetic arm to augment the command signal. The challenge is developing physiologically appropriate sensory feedback of multiple channel information, which is required in dexterous prehensile tasks.

Ephraim PL, Dillingham TR, Sector M, Pezzin LE, MacKenzie EJ. Epidemiology of limb loss and congenital limb deficiency: A review of the Literature. Arch. Phys. Med. Rehabil. 84, 747–761 (2003). Smith DG, Michael JW, Bowker JH (Ed.). Atlas of Amputations and Limb Deficiencies: Surgical, Prosthetic, and Rehabilitation Principles. 3rd Ed., AAOS, Rosemont, IL, USA (2004). A bible for prosthetic studies. Atkins DJ, Heard DCY, Donovan WH. Epidemiologic overview of individuals with upper-limb loss and their reported research priority. J. Prosthet. & Orthot. 8, 2–11 (1996). Gaine WJ, Smart C, Bransby-Zachary M. Upper limb traumatic amputees: Review of prosthetic use. J. Hand Surg. (Br ) 22B(1), 73–76 (1997). Davidson J. A comparison of upper limb amputees and patients with upper limb injuries using the Disability of the Arm,

Shoulder and Hand (DASH). Disabil. Rehabil. 26(14/15), 917–923 (2004). 10

Parker PA, Scott RN. Myoelectric control of prostheses. Crit. Rev. Biomed. Eng. 13, 283–310 (1986).

11

Basmajian JV, DeLuca CJ. Muscle Alive. 5th Ed, Williams & Wilkes, Baltimore, MD, USA (1985).

12

Muzumdar A (Ed.). Powered upper limb prostheses: Control, implementation and clinical application. Springer, Berlin, Germany (2004).

13

Hudgins B, Parker P, Scott RN. A New strategy for multifunction myoelectric control. IEEE Trans. Biomed. Eng. 40(1), 82–94 (1993).

14

Englehart K, Hudgins B, Parker PA, Stevenson M. Classification of the myoelectric signal using time-frequency based representations. Med. Eng. Phys. 2, 431–438 (1999).

Expert Rev. Med. Devices 4(1), (2007)

Neural machine interfaces for multifunctional powered upper limb prostheses

15

16

17

18

19

20

Chu JU, Moon I, Mun MS. A real-time EMG pattern recognition based on linear-nonlinear feature projection for multifunction myoelectric hand. Proceedings of 2005 IEEE 9th Int. Conf. on Rehabil. Robotics. Chicago, IL, USA, 295–298 (2005).

Peleg D, Braiman E, Yom-Tov E, Inbar GF. Classification of finger activation for use in a robotic prosthesis arm. IEEE Trans. Neural Syst. Rehabil. Eng. 10(4), 290–294 (2002). Karlik B, Tokhi MO, Alci M. A fuzzy clustering neural network architecture for multifunction upper-limb prosthesis. IEEE Trans. Biomed. Eng. 50(11), 1255–1261 (2003). Ajiboye AB, Weir RFff. A heuristic fuzzy logic approach to EMG pattern recognition for multifunctional prosthesis control. IEEE Trans. Neural Syst. Rehabil. Eng. 13(3), 280–291 (2005). Al-Assaf Y, Al-Nashash H. Surface myoelectric signal classification for prostheses control. J. Med. Eng. Technol. 29(5), 203–207 (2005). Chan ADC, Englehart KB. Continuous myoelectric control for powered prostheses using hidden Markov models. IEEE Trans. Biomed. Eng. 52(1), 121–124 (2005).

22

Huang Y, Englehart KB, Hudgins B, Chan ADC. A gaussian mixture model based classification scheme for myoelectric control of powered upper limb prostheses. IEEE Trans. Biomed. Eng. 52(11), 1801–1811 (2005).

24

•

25

Englehart K, Hudgins B. A robust, real-time control scheme for multifunction myoelectric control. IEEE Trans. Biomed. Eng. 50(7), 848–858 (2003).

27

Farrell TR, Weir RFff, Heckathorne CW. The effect of controller delay on box and block test performance. Conf. Proceedings of the Univ. of New Brunswick’s MyoElectric Controls / Powered Prosthetics Symposium (MEC’05). New Brunswick, Canada, 244–247 (2005).

Petterson PE, Anderson M. The use of self organizing maps to evaluate myoelectric signals. Biomed Sci Instrum. 35,147–152 (1999).

21

23

26

Bu N, Fukuda O, Tsuji T. EMG-based motion discrimination using a novel recurrent neural network. J. Intelligent Infor. Syst. 21(2), 113–125 (2003). Englehart K, Hudgins B, Parker P. Multifunction control of prostheses using the myoelectric signal. Chapter 5, In: Intelligent Systems and Technologies in Rehabilitation Engineering, Teodorescu H-N , Jain LC (Ed), CRC Press, Boca Raton, FL, USA, 153–208 (2000). Arguably the best article relating pattern recognition for prosthetic control. The book is also recommended. Zecca M, Micera S, Carrozza MC, Dario P. Control of multifunctional prosthetic hands by processing the electromyographic signal. Crit. Rev. Biomed. Eng. 30(4–6), 459–485 (2002).

www.future-drugs.com

28

29

30

Lock BA, Englehart K, Hudgin B. Real-time myoelectric control in a virtual environment to relate usability vs. Accuracy. Conf. Proceedings of the Univ. of New Brunswick’s MyoElectric Controls / Powered Prosthetics Symposium (MEC’05). New Brunswick, Canada, 122–126 (2005). Soares A, Andrade A, Lamounier E, Carrijo R. The development of a virtual myoelectric prosthesis controlled by an emg pattern recognition system based on neural networks. J. Intelligent Infor. Syst. 21(2), 127–141 (2003). Sebelius FCP, Rosén BN, Lundborg GN. Refined myoelectric control in below-elbow amputees using artificial neural networks and a data glove. J. Hand Surg. 30A(4), 780–789 (2005).

31

Fukuda O, Tsuji T, Kaneko M, Otsuka A. A human-assisting manipulator teleoperated by EMG signals and arm motions. IEEE Trans. Robotics Automation 19(2) 210–222 (2003).

32

Kuruganti U, Hudgins B, Scott RN. Twochannel enhancement of a multifunction control system. IEEE Trans. Biomed. Eng. 42(1), 109–111 (1995).

33

34

35

36

Cote J, Mathieu PA. Mapping of the human upper arm muscle activity with an electrode matrix. Electromyogr. Clin. Neurophysiol. 40(4), 215–223 (2000). Weir RFff, Troyk PR, DeMichele G, Kuiken T. Implantable Myoelectric Sensors (IMES) for upper-extremity prosthesis control – preliminary work. Proceedings of the 25th Annual International Conference of the IEEE EMBS, Cancun, Mexico, 1562–1565 (2003). Weir RFff, Troyk PR, DeMichele G, Lowery M, Kuiken T. Implantable myoelectric sensors (IMES). Conf. Proceedings of the Univ. of New Brunswick’s MyoElectric Controls / Powered Prosthetics Symposium (MEC’05). New Brunswick, Canada, 93–97 (2005). Loeb G. BIONTM system for distributed neural prosthetic interfaces. Med. Eng. Phys. 23(1), 9–18 (2001).

37

Arcos I, Davis R, Fey K, et al. Second-generation microstimulator. Artif. Organs 26(3), 228–231 (2002).

38

Stykov NS, Lowery MM, Heckman CJ, Taflove A, Kuiken TA. Recording Intramuscular EMG signals using surface electrodes. Proceedings of the 2005 IEEE 9th Int Conf. on Rehabil. Robotics, Chicago, IL, USA, 291–294 (2005)

39

Kuiken T. Consideration of nerve-muscle grafts to improve the control of artificial arms. Technol. Disabil. 15, 105–111 (2003).

40

Kuiken TA, Dumanian GA, Lipschutz RD, Miller LA, Stubblefield KA. The use of targeted muscle reinnervation for improved myoelectric prosthesis control in a bilateral shoulder disarticulation amputee. Prosthet. Orthot. Int. 28, 245–253 (2004).

41

Lipschutz RD, Miller LA, Stubblefield KA, Dumanian G, Phillips ME, Kuiken TA. Transhumeral level fitting and outcomes following targeted hyper-reinnervation nerve transfer surgery. Conf. Proceedings of the Univ. of New Brunswick’s MyoElectric Controls / Powered Prosthetics Symposium (MEC’05). New Brunswick, Canada, 2–5 (2005).

42

Miller LA, Lipschutz RD, Stubblefield KA, Kuiken TA. Fitting and outcomes of a bilateral shoulder disarticulation amputee following targeted hyper-reinnervation nerve transfer surgery. Conf. Proceedings of the Univ. of New Brunswick’s MyoElectric Controls / Powered Prosthetics Symposium (MEC’05). New Brunswick, Canada, 6–10 (2005).

43

Kyberd PJ, Chappell PH. The Southampton Hand: an intelligent myoelectric prosthesis. J. Rehabil. Res. Dev. 31(4), 326–334 (1994).

44

Light CM, Chappell PH, Hudgins B, Engelhart K. Intelligent multifunction myoelectric control of hand prostheses. J. Med. Eng. Technol. 26(4), 139–146 (2002).

45

Okuno R, Yoshida M, Akazawa K. Compliant grasp in a myoelectric hand prosthesisIEEE Eng. Med. Biol. Mag. 24(4), 48–56 (2005).

46

Horch KW, Dhillon GS (Ed.). Neuroprosthetics theory and practice: Theory and practice. World Scientific Publishing, Singapore, (2004). A complete source of information on neuroprosthesis.

• 47

Lawrence SM, Dhillon GS, Horch KW. Fabrication and characteristics of an implantable, polymer-based, intrafascicular electrode. J. Neurosci. Meth. 131, 9–26 (2003).

51

Ohnishi, Weir & Kuiken

48

49

50

Lawrence SM, Dhillon GS, Jensen W, Yoshida K, Horch KW. Acute peripheral nerve recording characteristics of polymer-based longitudinal intrafascicular electrodes. IEEE Trans Neural Syst. Rehabil. Eng. 12(3), 345–348 (2004).

59

Lawrence SM, Larsen JO, Horch KW, Riso R, Sinkjaer T. Long-term biocompatibility of implanted polymer-based intrafascicular electrodes. J. Biomed. Mater. Res. 63(5), 501–506 (2002).

60

Branner A, Stein RB, Fernandez E, Aoyagi Y, Normann RA. Long-term stimulation and recording with a penetrating microelectrode array in cat sciatic nerve. IEEE Trans. Biomed. Eng. 51(1), 146–157 (2004).

51

Branner A, Normann RA. A multielectrode array for intrafascicular recording and stimulation in sciatic nerve of cats. Brain Res. Bull. 51(4), 293–306 (2000).

52

Donaldson NDN, Zhou L, Perkins TA, Munih M, Haugland M, Sinkjaer T. Implantable telemeter for long-term electroneurographic recordings in animals and humans. Med. Biol. Eng. Comput. 41, 654–664 (2003).

53

54

55

56

Crampon MA, Brailovski V, Sawan M, Trochu F. Nerve cuff electrode with shape memory alloy armature: design and fabrication. Biomed. Mater. Eng. 12, 397–410 (2002). De Lucca CJ. Control of upper-limb prostheses: a case for neuroelectric control J. Med. Eng. Technol. 2(2), 57–61 (1978). Lichtenberg BK, De Luca CJ. Distinguishability of functionally distinct evoked neuroelectric signal on the surface of a nerve. IEEE Trans. Biomed. Eng. 26(4), 228–237 (1979). Kovacs GTA, Storment CW, Rosen JM. Regeneration microelectrode array for peripheral nerve recording and stimulation. IEEE Trans. Biomed. Eng. 39(9), 893–902 (1992).

57

DiLorenzo DJ, Edell DJ, Koris MJ, Riso RR. Chronic intraneural electrical stimulation for prosthetic sensory feedback. Proceedings of the 1st Int. IEEE EMBS Conf. on Neural Eng. Capri Island, Italy, 116–119 (2003).

58

Stieglitz T, Ruf HH, Gross M, Schuettler M, Meyer J-UA. Biohybrid system to interface peripheral nerves after traumatic lesions: design of a high channel sieve electrode. Biosens. Bioelectron. 17, 685–696 (2002).

52

61

Lago N, Ceballos D, Rodriguez FJ, Stieglitz T, Navarro X. Long term assessment of axonal regeneration through polyimide regenerative electrodes to interface the peripheral nerve. Biomaterials 26(14), 2021–2031 (2005). Edell DJ. A peripheral nerve information transducer for amputees: long-term multichannel recordings from rabbit peripheral nerves. IEEE Trans. Biomed. Eng. 33(2), 203–214 (1986). Navorro X, Krueger TB, Lago N, Micera S, Stieglitz T, Dario P. A critical review of interfaces with the peripheral nervous system for the control of neuroprostheses and hybrid bionic systems. J. Peripheral Nervous Syst. 10, 229–258 (2005).

62

Dhillon GS, Lawrence SM, Hutchinson DT, Horch KW. Residual function in peripheral nerve stumps of amputees: Implications for neural control of artificial limbs. J. Hand Surg. 29A(4), 605–615 (2004).

63

Dhillon GS, Krüger TB, Sandhu JS, Horch KW. Effects of short-term training on sensory and motor function in severed nerves of long-term human amputees. J. Neurophysiol. 93, 2625–2633 (2005).

64

Warwick K, Gasson M, Hutt B, et al. The application of implant technology for cybernetic systems. Arch. Neurol. 60(10), 1369–1373 (2003).

65

Gasson M, Hutt B, Goodhew I, Kyberd P, Warwick K. Invasive neural prosthesis for neural signal detection and nerve stimulation. Int. J. Adaptive Control Signal Proc. 19(5), 365–375 (2005).

66

Wolpaw JR, Birbaumer N, McFarland DJ, Pfurtscheller G, Vaughan TM. Brain–computer interfaces for communication and control. Clin. Neurophysiol. 113, 767–791 (2002). Arguably the best paper available relating the brain–computer interface.

• 67

Wolpaw JR, Birbaumer N, Heetderks WJ et al. Brain–computer interface technology: a review of the first international meeting. IEEE Trans. Rehabil. Eng. 8(2), 164–173 (2000).

68

Vaughan TM, Heetderks WJ, Trejo LJ et al. Brain-computer interface technology: a review of the Second International Meeting. IEEE Trans. Neural Syst. Rehabil. Eng. 11(2), 94–109 (2003).

69

Nicolelis MA. Brain-machine interfaces to restore motor function and probe neural circuits. Nat. Rev. Neurosci. 4(5), 417–422 (2003).

70

Donoghue JP. Connecting cortex to machines: recent advances in brain interfaces. Nature Neurosci. 5, 1085–1088 (2002).

71

Friehs GM, Zerris VA, Ojakangas CL, Fellows MR, Donoghue JP.Brain-Machine and Brain-Computer Interfaces. Stroke 35 (Suppl. I), 2702–2705 (2004).

72

Lauer RT, Peckham PH, Kilgore KL, Heetderks WJ. Applications of cortical signals to neuroprosthetic control: a critical review. IEEE Trans. Rehabil. Eng. 8(2), 205–208 (2000).

73

Pfurtscheller G, Guger C, Müller G, Krausz G, Neuper C. Brain oscillations control hand orthosis in a tetraplegic. Neurosci. Lett. 292, 211–214 (2000).

74

Pfurtscheller G, Neuper C, Müller GR et al. Graz-BCI: State of the art and clinical applications. IEEE Trans. Neural Syst. Rehabil. Eng. 11(2), 177–180 (2003).

75

Abboudi RL, Glass CA, Newby NA, Flint JA, Craelius W. A biomimetic controller for multifinger prosthesis. IEEE Trans. Rehabil. Eng. 7(2), 121–129 (1999).

76

Curcie DJ, Flint JA, Craelius W. Biomimetic finger control by filtering of distributed forelimb pressures. IEEE Trans. Neural Syst. Rehabil. Eng. 9(1), 69–75 (2001).

77

Silva J, Heim W, Chau T. A self-contained, mechanomyography-driven externally powered prosthesis. Arch. Phys. Med. Rehabil. 86, 2066–2070 (2005).

78

Gregori B, Galie E, Accornero A. Surface electromyography and mechanomyography recording: a new differential composite probe. Med. Biol. Eng. Comput. 41, 665–669 (2003).

79

Heath GH. Control of proportional grasping using a myokinemetric signal. Technol. Disabil. 15, 73–83 (2003).

80

Davalli A, Sacchettia R, Fanin S, Avanzolini G, Urbano E. Biofeedback for upper limb myoelectric prostheses. Technol. Disabil. 13, 161–172 (2000).

81

Lundborg G, Rosen B, Lindberg S. Hearing as substitution for sensation: a new principle for artificial sensibility. J. Hand Surg. (Am) 24(2), 219–224 (1999).

82

Kuiken TA, Sensinger JW, Farrell TR, Weir RFff. Phantom limb sensory f eedback through nerve transfer surgery. Conf. Proceedings of the Univ. of New Brunswick’s MyoElectric Controls/Powered Prosthetics Symposium (MEC’05). New Brunswick, Canada, 25–29 (2005).

83

Lundborg G, Rosen B. Sensory substitution in prosthetics. Hand Clin. 17(3), 481–488 (2001). Comprehensive review on sensory substitution.

••

Expert Rev. Med. Devices 4(1), (2007)

Neural machine interfaces for multifunctional powered upper limb prostheses

84

•• 85

Riso RR. Strategies for providing upper extremity amputees with tactile and hand position feedback -moving closer to the bionic arm. Technol. Health Care 7, 401–409 (1999) The best paper for sensory neural interface. Dhillon GS, Horch KW. Direct neural sensory feedback and control of a prosthetic arm. IEEE Trans. Neural Syst. Rehabil. Eng. 13(4), 468–472 (2005).

86

Romo R, Hernandez A, Zainos A, Broody CD, Lemus L. Sensing without touching: psychophysical performance based on cortical microstimulation. Neuron 26, 273–278 (2000).

87

Simpson DC. The choice of control sustem for the multi-movement prosthesis: Extended physiological proprioception (EPP). In: The control of upper-extremity prostheses and orthoses, (Herberts P et al. Eds), Springfield, Illinois, C.C Thomas, 146–150 (1974).

88

Maruishi M,Tanaka Y, Muranaka H et al. Brain activation during manipulation of the myoelectric prosthetic hand: a functional magnetic resonance imaging study. NeuroImage 21, 1604–1611 (2004).

www.future-drugs.com

Websites 101

Biomedical Engineer, Northwestern University Prosthetic Research Laboratory 345 E. Superior, RM1441, Chicago, Illinois 60611, USA Tel.: +1 312 238 6500; +1 312 238 6512 Fax: +1 312 238 6510 [email protected]

OTTO BOCK HealthCare, Sensor Hand® SPEED www.ottobockus.com/products/upper_lim b_prosthetics/myoelectric_hands_ sensorhand.asp Accessed June 2006

Affiliations •

Kengo Ohnishi Visiting Professor, Northwestern University Prosthetic Research Laboratory Research Associate, Oita University, Department of Welfare Engineering, Faculty of Engineering, 700 Dannoharu, Oita, 8701192, Japan Tel.: +81 97 554 7771 Fax: +81 97 554 7507 [email protected]

•

Richard F Weir Research Scientist, Jesse Brown VA Medical Center - Lakeside CBOC Research Assistant Professor, Northwestern University, Department of Physical Medicine & Rehabilitation, Feinberg School of Medicine Department of Biomedical Engineering, McCormick School of Engineering & Applied Science

•

Todd A Kuiken Director, Neural Engineering Center for Artificial Limbs, Rehabilitation Institute of Chicago Associate Professor, Northwestern University Department of Physical Medicine & Rehabilitation, Feinberg School of Medicine Department of Biomedical Engineering, McCormick School of Engineering & Applied Science 345 E. Superior St. RM1309, Chicago, IL, USA Tel.: +1 312 238 8072 Fax: +1 312 238 1166 [email protected]

53