TMR & Pattern Recognition
Paired with conventional prosthetic technologies, TMR can significantly improve prosthesis control for individuals with transhumeral or shoulder disarticulation amputations. Ongoing research is focused on extending the surgical technique to individuals with transradial amputations and lower limb amputations.
Other significant priorities include developing prosthetic technologies to provide further functional improvements, and designing improved control paradigms to capitalize on the additional neural control information made available by TMR.
Improved Control Strategies
Conventional EMG control techniques are based on the magnitude of EMG signals and give users basic control of powered prostheses.
However, these techniques are relatively primitive; they use only a fraction of the available control information and are confounded by muscle cross talk—EMG signals from other adjacent muscles.
Consequently, these techniques require specific electrode placement, achieved through trial and error, to optimize control. This can be challenging in individuals with high level amputations, as they need to control several degrees of freedom but have few suitable control sites available.
In addition, because few residual muscles remain after above-elbow amputation, the user must often control several degrees of freedom using one muscle pair. This means that the user must use unintuitive contractions to control the prosthesis, for example, using the biceps and triceps to control the hand and wrist. The user must also switch the prosthesis, for example, from hand to elbow mode. Mode-switching is often done by co-contraction.
Pattern recognition provides an exciting new way to control a myoelectric prosthesis. As shown in Figure 1, different attempted movements generate distinct, characteristic patterns of muscle activation, which in turn, generates unique EMG patterns—like an electrical fingerprint.
Figure 1: EMG signal amplitudes from 115 high-density electrodes generated when a TMR patient attempted the different movements shown (colors represent differnet EMG signal intensities, according to the key at right). These maps demonstrate clear differences in muscle activation patterns with different intended movements. This patient had a sholder disarticulation ampuration, so nerves were transferred to muscules in the chest.
Computer algorithms can learn to recognize and distinguish these patterns; once trained, the algorithm can then decipher what the user intends to do and command the prosthesis to perform that movement.
Pattern recognition makes control of a prosthesis more intuitive, or natural: the user simply has to try to make the desired movement with their residual limb, and the prosthesis responds with the correct movement.
In addition, users do not have to switch the prosthesis between modes to control different degrees of freedom, so control is seamless.
Pattern recognition technology also does not require specific electrode placement, which makes clinical fitting simpler and faster.
Although pattern recognition works for people who have not had TMR, an individual who has undergone TMR may benefit even more from pattern recognition, because TMR creates additional EMG signals that can be incorporated into the patterns, creating a more detailed pattern. TMR also allows access to the rich neural information that is carried by transferred brachial plexus nerves for control of the arm, hand, and digits.
Recently, three TMR patients were fitted with a commercially available pattern recognition system through Coapt LLC. One of these individuals had shoulder disarticulation amputation, and one had transhumeral amputation. The third individual had bilateral amputations (left shoulder disarticulation, right transhumeral), and was the first person to undergo bilateral TMR.
Full disclosure: Coapt LLC was launched in 2012 and has a technology transfer and license agreement with the Rehabilitation Institute of Chicago for the development of certain control technologies. Several current and former members of the Center for Bionic Medicine at RIC have management and ownership interests in Coapt LLC.
Combining TMR with pattern recognition for other amputee populations
So far, TMR has been performed primarily on individuals with transhumeral or shoulder disarticulation amputations. However, as research progresses, TMR may become available for individuals with transradial and transfemoral amputations—dramatically increasing the number of individuals who can benefit from TMR.
TMR for Individuals with Transradial amputation
Transradial amputation is the most common major upper limb amputation, affecting several thousand individuals in the United States. Although transradial amputees have more residual muscles than higher level amputees, these individuals have at most only two residual muscles to control a prosthetic hand using conventional control. Using pattern recognition, transradial amputees cannot perform as many hand grasp movements as individuals with higher level amputations who have had TMR [1, 2] (Fig. 1) because transferred nerves carry control information for intrinsic hand finger, and thumb muscles, which is not available to individuals with a transradial amputation.
Figure 1: Task completion rates for 5 transradial
amputees performing 5 hand grasp patterns and
for 5 TMR patients performing 3 hand grasp patterns.
When three hand grasps are compared, task
completion rates were significantly higher for
TMR patients. (Click image to enlarge)
Performing transradial TMR, in which residual nerves from intrinsic hand muscles are transferred to remaining forearm muscles, provides individuals with intuitive control over prosthetic wrist and hand movements, and potentially allows for more intricate grasping patterns. This is significant given that advanced, multifunction hands and fingers are now commercially available. Thus transradial TMR could potentially benefit thousands of individuals .
Currently, we are evaluating ways to quantify the control information provided by intrinsic hand muscles to guide future transradial TMR procedures .
TMR for individuals with lower limb amputation
Nearly two-thirds of all amputations occur in the lower-limb—primarily due to dysvascular disease . Individuals with lower limb amputation are in great need of robust, safe prostheses. Microprocessor-controlled powered leg prostheses have recently become available. These devices are controlled either through mechanical sensors placed on the prosthesis, external remote key fobs, or an instrumented orthotic on the sound side.
Figure 2: Control of non-weight-bearing movement by
two TMR patients (TMR 1 and TMR 2) and 4
non-TMR subjects. (Click image to enlarge)
EMG signals produced in the residual limb may provide improved control of such legs. EMG signals can be combined with information from mechanical sensors; the resulting control system is more accurate and more responsive to different ambulation modes, such as stair climbing or level-ground walking. EMG control also facilitates control of non-weightbearing activities (Fig. 2), which, for example, would facilitate dressing or adjusting the position of the prosthetic leg for comfort.
TMR may further enhance control by providing access to the neural commands intended for the distal muscles of the leg lost by amputation. Many muscles in the residual leg are suitable for nerve transfer .
Although TMR for lower limb prosthesis control has not been performed, the procedure has been performed to prevent or treat neuromas in individuals with lower limb amputation.
The following case study published in the New England Journal of Medicine demonstrates the use of EMG from reinnervated and natively innervated muscles to provide enhanced control of a powered prosthetic leg . In 2009, a 31-year-old man underwent a knee-disarticulation surgery amputation approximately 36 hours after a motorcycle accident. During surgery, two nerve transfers were performed to prevent neuroma formation—a procedure analogous to TMR.
EMG patterns after reinnervation and the subject’s control of a powered prosthesis using pattern recognition were evaluated. Using EMG signals from natively innervated and surgically innervated (i.e. via TMR) residual thigh muscles improved the patient’s control of his robotic leg. Specifically, (1) the reinnervated hamstring muscles generated strong EMG signals; (2) a unique stride pattern for each ambulation mode was produced from EMG data; and (3) the classification accuracy of attempted movements was 96.0%.
Using this control system, this individual was able to climb 103 flights of stairs to the top of the Willis Tower in Chicago.
Read an interview with the subject.
Other media coverage:
First mind-controlled bionic leg a 'groundbreaking' advance (NBC News)
RIC Will Unveil World’s First Neural-Controlled Bionic Leg at Fourth Annual Skyrise Chicago Event (RIC news coverage and video)
1. Kuiken TA, Li G, Lock BA, et al. Targeted muscle reinnervation for real-time myoelectric control of multifunction artificial arms. JAMA : the journal of the American Medical Association. Feb 11 2009;301(6):619-628.
2. Li G, Schultz AE, Kuiken TA. Quantifying pattern recognition-based myoelectric control of multifunctional transradial prostheses. IEEE transactions on neural systems and rehabilitation engineering : a publication of the IEEE Engineering in Medicine and Biology Society. Apr 2010;18(2):185-192.
3. Dillingham TR, Pezzin LE, MacKenzie EJ. Limb amputation and limb deficiency: epidemiology and recent trends in the United States. Southern medical journal. Aug 2002;95(8):875-883.
4. Li G, Schultz AE, Kuiken TA. Quantifying pattern recognition-based myoelectric control of multifunctional transradial prostheses. IEEE transactions on neural systems and rehabilitation engineering : a publication of the IEEE Engineering in Medicine and Biology Society. Apr 2010;18(2):185-192.
5. Ziegler-Graham K, MacKenzie EJ, Ephraim PL, Travison TG, Brookmeyer R. Estimating the prevalence of limb loss in the United States: 2005 to 2050. Archives of physical medicine and rehabilitation. Mar 2008;89(3):422-429.
6. Agnew SP, Schultz AE, Dumanian GA, Kuiken TA. Targeted reinnervation in the transfemoral amputee: a preliminary study of surgical technique. Plastic and reconstructive surgery. Jan 2012;129(1):187-194.
7. Hargrove LJ, Simon AM, Young AJ, et al. Robotic leg control with EMG decoding in an amputee with nerve transfers. The New England journal of medicine. Sep 26 2013;369(13):1237-1242.
Advanced Prosthetic Systems
Despite steady advances in technology, myoelectric prostheses are still too heavy for the average user, especially women and children; and all prosthesis users must deal with burdensome sockets, which can irritate the skin and cause discomfort. These challenges often result in rejection of the device altogether.
As TMR allows individuals to control a greater number of prosthetic movements, the industry has responded by designing lighter prostheses with more control inputs, including the Rehabilitation Institute of Chicago’s lightweight modular arm that provides six degree of freedom and is suitable for individuals as small as a 25th percentile woman (or 50th percentile child). OttoBock’s DynamicArm® and several new motorized leg systems currently under development. Despite these advancements, designing lighter prostheses with robust neural interfaces remains an important priority.
Intramuscular (implantable) EMG sensors
While surface electrodes are inexpensive and non-invasive, they are also prone to muscle cross talk – EMG signal contributions that originate from muscles but are not directly under the electrodes and electrode shift (movement of electrodes with respect to the skin surface during donning or use of the prosthesis). Cross talk and/or electrode shift often compromise the success of control algorithms. In contrast, intramuscular EMG sensors may provide high-quality, cross talk–free measures of activation. Further, intramuscular (implantable) EMG sensors enable researchers to attempt new myoelectric control strategies that may not be feasible with surface EMG.
Intramuscular EMG sensors:
- Allow the recording of muscles deep in the forearm, such as the supinator muscle, which would be nearly impossible using conventional recording methods.
- Prevent the shifting of electrodes during donning / doffing. Intramuscular EMG sensors may therefore provide a more stable interface and require less frequent training of the control system.
- Allow recording from inside the muscle, which opens up the skin surface for potential application of sensory feedback.
Further reading on intramuscular EMG sensors:
A Comparison of the Effects of Electrode Implantation and Targeting on Pattern Classification Accuracy for Prosthesis Control (IEEE)
Development of an Implantable Myoelectric Sensor for Advanced Prosthesis Control (Artificial Organs)
Simultaneous and Proportional Force Estimation in Multiple Degrees of Freedom From Intramuscular EMG (IEEE)
Intramuscular EMG after targeted muscle reinnervation for pattern recognition control of myoelectric prostheses (IEEE)
Levi J. Hargrove, PhD, received BSc, MSc, and PhD degrees in electrical engineering from the University of New Brunswick, Fredericton, New Brunswick, Canada. Since 2008, Dr Hargrove is the Director of the Neural Engineering for Prosthetics and Orthotics Lab at the Center for Bionic Medicine, Rehabilitation Institute of Chicago. He is also a Research Assistant Professor in the Departments of Physical Medicine and Rehabilitation and in Biomedical Engineering at Northwestern University. His research interests include pattern recognition, biological signal processing, and myoelectric control of powered prostheses. Dr. Hargrove is a member of the Association of Professional Engineers and Geoscientists of New Brunswick.
Blair A. Lock, MS, received BS and MS degrees in electrical engineering and a diploma in technology management and entrepreneurship from the University of New Brunswick, Fredericton, NB, Canada. Mr. Lock is a collaborator with the Center for Bionic Medicine, Rehabilitation Institute of Chicago. He is also managing partner of Coapt LLC, a company specializing in upper-extremity control technology. His research interests include pattern recognition for improved control of powered prostheses and user experience in rehabilitation technologies. Mr. Lock is a registered professional engineer with the Association of Professional Engineers and Geoscientists of New Brunswick, Canada.
Lauren H. Smith received a BS in biomedical engineering from Northwestern University. She is currently working toward both an MD degree and a PhD degree in biomedical engineering at Northwestern University. She conducts her doctoral research at the Center for Bionic Medicine under the mentorship of Drs. Todd Kuiken and Levi Hargrove. Her work focuses on using implantable recording devices to improve control of prosthetic arms. Her broader research interests include neural interface, bioelectric signal analysis, and clinical neurophysiology.