Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2021 Mar 1.
Published in final edited form as: IEEE Robot Autom Mag. 2019 Nov 20;27(1):77–86. doi: 10.1109/mra.2019.2949688

A Myoelectric Postural Control Algorithm for Persons With Transradial Amputations: A Consideration of Clinical Readiness

Jacob L Segil 1, Rahul Kaliki 2, Jack Uellendahl 3, Richard F ff Weir 4
PMCID: PMC7269158  NIHMSID: NIHMS1578578  PMID: 32494115

Abstract

Background

The bottleneck in upper limb prosthetic design is the myoelectric control algorithm. Here we studied the clinical readiness of the myoelectric postural control algorithm in a laboratory setting with two trans-radial amputees using a commercially available prosthetic limb system.

Technique

The postural control algorithm was integrated into prosthetic limb systems using standard of care components. A comparison between a commercial state of the art system (the i-limb revolution state-based myoelectric controller) and the postural controller was performed with two people with trans-radial amputation using a self-contained prosthesis system.

Discussion

The performance using the i-limb revolution state-based controller versus the postural controller was mixed based on the Southampton Hand Assessment Procedure. The SHAP scores indicate that the postural controller with i-limb revolution provided an average of 66% of hand function compared to an intact limb. Future work will study the advantages of the postural control algorithm in everyday use.

I. Introduction

It is widely understood that current technology in upper limb prosthetics does not fulfill all of the needs of persons with upper limb amputation [1]. Device abandonment is as high as 50% [1] and myoelectric prosthetic devices are not favored compared to older, cumbersome body-powered technology [2]. Much effort in the field is spent on the electromechanical design of multifunctional prosthetic hands [3] while the parallel challenge of advanced myoelectric control techniques has seen less translation into commercially available technologies. The mechanical design of many prosthetic hands today offers far more functionality than is available to the user due to the control methodologies implemented with these devices [4]. In other words, the myoelectric control algorithm is the bottleneck in the prosthetic limb system. The elimination of this bottleneck should enhance users’ abilities and hopefully lead to better quality of life for persons with upper limb amputation.

The first commercially available multi-function prosthetic hands used state-based direct myoelectric control techniques that are still used today. Direct control uses triggers, or co-contractions of two electromyographic (EMG) signals, to sequentially command different grasps. Then, the user commands a velocity-based open/close signal using the extension/flexion EMG sites. Users consider muscle triggers such as a co-contraction to control multiple grip patterns or movements slow, cognitively demanding and unintuitive [5–7]. Furthermore, during sustained muscle contraction, there are shifts in the amplitude and frequency of the EMG signal, which demonstrate the onset of fatigue [8]. Fatigue can cause problems in the control of myoelectric prosthetic devices [9]. The switching mechanisms outlined above require more muscle activation (particularly hold open, which requires sustained muscle contraction) and conscious effort. As a result, users can fatigue their residual muscles much more quickly [10].

Our own preliminary work has shown that these EMG triggers are inconsistent and can be frustrating to use. Following is a quote from an interview with amputee S.M. elaborating on his current myoelectric prosthesis: “It is capable of four different grip patterns, and I’ve only been able to successfully actuate two of them. The double impulse and the triple impulse, for some reason, I’ve tried every, every arrangement in the set up to try to make that happen and I just can’t seem to do it, except at times involuntary. […] So, if I want to get the finger point on that, it can take a while, or it may not do what I ask it to do.” [11]. Examples like this show the difficulty users can experience when switching grips.

A new control method, called pattern recognition, is emerging while several hurdles remain. Many researchers (including ourselves) have turned to pattern recognition of multichannel myoelectric signals in order to develop more intuitive control of advanced prostheses [12–15]. Pattern recognition algorithms seek to correlate patterns of surface EMG activity with a given intended movement command. Correlation is determined by calibrating a machine-learning algorithm with labeled training examples in the form muscle activity recorded while the user holds a static posture. Because these patterns are representative of natural behaviors prior to amputation, control of the prosthesis via pattern recognition is intuitive and potentially increases the number of controllable DOFs. There has been significant progress in developing these algorithms over the past two decades [15–17]. This work has resulted in the launch of the first pattern recognition system on the market from Coapt, LLC and now from Infinite Biomedical Technologies, LLC. However, this system is still limited to operating 1 DOF in the hand (no grip patterns) and only 3 DOFs total (hand, wrist, and elbow).

The most significant challenge for pattern recognition algorithms is that they require highly consistent and noise-free EMG signals. This is particularly true as the number of degrees of freedom (DOF) in prosthetic hands increase. Moving from one grip pattern to another may result in overlapping muscle activity patterns, making the classifier inaccurate. Upper limb amputees experience this issue in a pronounced way especially due to the loss of sensory feedback from the amputated limb and motor cortex reorganization [18, 19]. Our own pioneering work in this area has shown that users can form more consistent and separable patterns with 10 days of diligent training with a visual feedback system [20, 21]. In addition, extrinsic factors such as electrode displacement, moving to positions outside of the initially trained position, and using the prostheses under varying loads [14, 22, 23] can all significantly degrade performance. Recalibration of the prosthesis can solve these issues; however, a recalibration can take several minutes and multiple daily recalibration session pose significant usability challenges. Thus, it would be highly preferable to develop a solution that can work without any calibration or extensive patient training.

In all cases, the various myoelectric control schemes provide different methods for the user to command the multi-functional prosthetic hand. In other words, these algorithms simply vary the sequence of steps to open/close the hand into the desired grasp/posture. The main differences among myoelectric control techniques are 1) the number of electrodes, 2) the requirement for device training, 3) intuitiveness of control, and 4) number of accessible grasps. The optimal myoelectric control scheme would 1) reduce the number of electrodes and associated circuit complexity, 2) eliminate the need to device training, 3) provide intuitive control, and 4) ensure all functional grasps are easily accessible. While this ideal control scheme is not available today, an alternative method to the state-based and pattern recognition systems used previously, namely the postural controller, is presented here.

This alternative type of myoelectric control uses a multi-dimensional direct control paradigm where the EMG is mapped into a two-dimensional domain which is then transformed to control a multi-dimensional prosthesis [23–26]. In this case, the postural controller is akin to using a joystick to control a computer cursor, a task humans are very capable of performing with high accuracy and precision [27–29]. This control strategy uses a position or velocity control scheme that relates EMG activity to the position of each digit using the joint angle transform (patent pending: PCT/US2014/040569). This allows users to seamlessly glide from one posture to the next, resulting in more dexterous use of multi-articulated hands. Practically, this means that patients will be able to learn this strategy quickly without excessive occupational therapy. Previous work has demonstrated the postural controller with able-bodied subjects [30] and trans-radial amputees [31]. However, we have not confirmed the utility of the postural control algorithm when using commercially available prostheses as well as clinically sound, self-suspending prosthetic sockets. Here we implemented the postural control algorithm on a commercially available prosthetic hand and compared it to the state-based myoelectric controller provided with the hand. A discussion follows concerning the clinically viability of the control algorithm and the clinical implications of this type of control technique on the patient and prosthetist alike.

II. Materials and Methods

A. Participant Details

Two subjects with trans-radial limb loss (2 men; average age of 45, 1 with traumatic limb loss, 1 with congenital limb absence) completed the experimental protocol in the Biomechatronics Development Laboratory. The Colorado Multiple Institution Review Board approved the study and both subjects provided written informed consent. Table 1 provides the subject demographic details including the type of limb loss, affected side, length of residual limb, age, typical prosthesis use, and experience with a multi-functional prosthetic hand. Subject S1 had extensive experience using multi-functional prosthetic hands; he owns an iLimb (Touch Bionics, Ltd.) and uses it on a regular basis. Subject S2 had minimal experience using a multi-functional prosthetic hand; he never has owned a multi-functional prosthesis and only tried them in his prosthetic clinic.

Table 1. –

Demographic data on two experimental subjects

Subject Type of Limb Loss Affected Side Length of Residual Limb Age Typical Type of Prosthesis Myoelectric Control Experience
S1 Amputation Left 2” below elbow 34 years Myoelectric prosthetic hand Extensive
S2 Congenital limb absence Right 5” below elbow 55 years Body-powered hook Minimal
Open in a new tab

B. Prosthetic Fitting

The prosthetic fitting for a transradial prosthesis using postural control algorithm is no different from those used for most myoelectric direct control prostheses. The postural controller simply requires the addition of a third electrode assembly in the socket. The location of the third electrode is on the ulnar side of the residual limb over the extensor carpi ulnaris. When fabricating a prosthetic socket, the prosthetist positions the electrodes to record signals associated with forearm flexion, extension and ulnar deviation. Once the prosthetist identifies suitable electrode sites, the prosthetist transfers the sites to the test socket. The prosthetist temporarily installs the electrodes in the test socket and then attaches ancillary components (batteries, etc.) for dynamic trial of the complete system. The prosthetist fabricates the definitive socket in the usual manner using the electrode dummies provided by the manufacturer. Finally, the prosthetist installs the additional controller electronics in the hollow area of the forearm or sandwiched between the socket and forearm shell in the case of a very long residual limb.

In this work, custom prosthetic sockets were fabricated using High Consistency Rubber (HCR) silicone. The prosthetist embedded fabrication dummies for the three electrode sites into the HCR. Although the fitted sockets were made of HCR silicone, the prosthetist performed fabrication trials to evaluate the fabrication process using flexible thermoplastics as well. Both materials required fabrication techniques that are similar to those used in contemporary clinical practice.

The only difference in fabricating a socket for postural control is the addition of a third electrode and the identification of a third not cross-coupled EMG site – using standard-of-care procedures to locate. This requirement did not pose any significant challenge in the two cases fitted. We assume that the addition of a third electrode assembly may become difficult for very short or small residual limbs since the available area to mount the electrode may be insufficient in these cases.

Myoelectric site selection was more difficult than dual site because the postural controller requires independence among the three electrode sites. Current myoelectric testing equipment typically has only two test electrodes. This makes simultaneous evaluation of three sites impossible with current equipment. The postural control algorithm user interface (as described below) visualizes all three myoelectric signals simultaneously to ensure proper prosthetic fitting.

Each prosthetic limb systems included the following components: 1) three electromyography electrodes, 2) an embedded microprocessor to communicate with the prosthetic hand, 3) a multi-articulated prosthetic hand with 6 degrees-offreedom (DoFs) including a 2 DoF thumb, 4) silicon glove, 5) prosthetic socket, and 6) batteries. The self-suspending prosthetic limb systems (Figure 1) used all commercially available technologies to ensure an accurate depiction of a prosthetic fitting for the postural controller in a clinical setting today.

Figure 1. –

Figure 1. –

Open in a new tab

Prosthetic limb systems produced for experimental sessions. The prosthesis included the iLimb Revolution prosthetic hands, electrodes, battery, and controller all embedded within a self-suspending test socket.

C. Myoelectric Controllers

The type of myoelectric control algorithm used during a given experimental session was the independent variable. We implemented two myoelectric control algorithms using the identical prosthetic limb system; the software was simply switched between experimental sessions. The first myoelectric controller (C1) is the commercial myoelectric control system provided in the i-limb™ revolution (Össur hf., Reykjavik, Iceland). The i-limb™ can be controlled various methods including gesture control, grip chips, or the use of buttons. However, we utilized only the myoelectric controller during this experiment. This control scheme is a state-machine controller where users switch grasp/posture states using certain “muscle triggers”. Then, the hand is open/closed into the specific grasp/posture using a velocity control paradigm. Many trigger options were available including 1) hold open – a held extension EMG signal of longer than 1 second, 2) co-contraction – simultaneous EMG signals from both the flexor and extensor site of a magnitude greater than a pre-determined threshold, 3) double impulse – a sequence of two large EMG signals within a certain time window. The accompanying biosim™ software (Össur hf., Reykjavik, Iceland) was used to setup the state machine where the three functional grasps required for the experimental (palmar prehension, lateral prehension, and tip prehension) were available using the preferred triggers for each subject. EMG gains and thresholds were also adjusted using the biosim software in order to create the most robust myoelectric control interface for each subject.

The second myoelectric controller (C2) is the postural controller as described previously by the authors in other work [26, 30–32]. In this experiment, we implemented the postural controller as a custom embedded program on the CORE operating platform provided by Infinite Biomedical Technologies. We rewrote the software in embedded C and we monitored the program using a custom graphical user interface developed in C# that communicated with the prosthesis using a Bluetooth communication protocol. In brief, the postural control algorithm transforms surface EMG signals into a joint angle array (or hand posture). The EMG signals are mapped to a postural control domain (see dashed lines in Figure 2) where the flexor digitorum EMG site was mapped to the eastern direction, the extensor digitorum EMG site was mapped to the northwest direction, and the ulnar deviation EMG site was mapped to the southwest direction. The instantaneous EMG activity was summed using vector addition to calculate a resultant vector (shown in red in Figure 2). The algorithm calculates the speed and direction for all six joints in the hand based on the magnitude and location of the resultant vector. A joint angle transform (a linear transform between the 2D resultant vector and 6D joint angle space) determines the instantaneous hand posture.

Figure 2. –

Figure 2. –

Open in a new tab

Postural control domain where each slice controls a different functional grasp/posture as indicated. The real-time EMG resultant vector (red) is a vector summation of the instantaneous EMG activity across the three electrodes (dashed black lines). The vector magnitude determines the speed of closure and the location of the vector determines the grasp type.

The postural control domain contains six functional grasps/postures including 1) palmar prehension (yellow), 2) opposition (light blue), 3) hand open (orange), 4) hook (green), 5) lateral prehension (blue), and tip prehension (purple). The user manipulates the instantaneous EMG signals in order to drive the resultant vector into any of the grip “slices”. When the vector is within a particular “slice”, then prosthesis switches into the pre-state of the grasp and begins to close to hand into the desired grasp/posture where the speed of closure is proportional to the magnitude of the EMG. No EMG activity caused no motion of the prosthesis. If the vector enters the 3) hand open (orange) domain, then the prosthesis opens/extends all joints simultaneously. This posture was necessary to map to a certain “slice” since the velocity control paradigm require a method to open the hand after closed into a grip. (For a lengthier discussion on the differences between position and velocity control using the postural control algorithm see Segil and Weir, 2015.) We detail the decision to use a velocity control methodology in this study in the Discussion section of this manuscript.

D. Prosthetic Limb System

This experiment used the i-limb revolution and the robo-limb prosthetic hands. (In fact, the robo-limb (Touch Bionics by Össur, Inc.) was used to implement the postural control algorithm. The i-limb revolution and the robo-limb are identical in prosthetic devices in terms of their mechanical design.) The robo-limb allows for direct communication with the actuators and is therefore necessary to perform custom control methodologies like the postural control algorithm. The i-limb revolution was used in conjunction with the direct control system (C1) and the robo-limb was used with the postural controller (C2). Both of these devices are mechanically identical; the robo-limb allows for custom actuation of the digits for experimental testing using a proprietary CAN bus protocol. This device is the first commercially available six degree-of-actuation (DoA, i.e. – number of actuators/motors) prosthetic hand. The prosthesis includes actuators in each digit and an actuator to drive the thumb abduction/adduction. The actuators do not include position encoding (see Discussion section for more details concerning this feature of the prosthesis). The hand closes in 1.2 seconds, weighs between 443g – 515g depending on the hand size and wrist configuration, and has a static fingercarrying load of 71 lbs. A silicon glove fit over the prosthetic hand to provide a high friction surface. A passive wrist component allowed for the rotation of the hand about the forearm by using the contra-lateral limb to position the hand in rotation.

The communication protocol of the robo-limb was critical to the integration of the CORE operating platform with the existing prosthesis. The CORE platform used a CAN bus to communicate with the individual motors within the robo-limb. A limited number of feed-forward commands were available including: 1) movement direction of motor and 2) movement speed of motor. The feedback information available on the CAN bus included 1) thumb abduction location, 2) digit status if stopped, moving, or stalled, 3) measured motor current draw. The output of the postural control algorithm in the CORE platform was translated into appropriate CAN bus commands to each digit (a speed and direction). The CORE controller monitored the ‘duration of motion’ in order to ensure each digit moved to the appropriate position for any given grasp. While unideal, this implementation of the postural control algorithm was a compromise between the algorithmic capabilities and the limitations of the commercially available prosthetic system.

E. Experimental Methods

Both subjects completed the same sequence of visits and experiments to the Biomechatronics Development Laboratory. The first visit entailed the prosthetic fitting and testing as described previously. Then, each subject attended two experimental sessions across two different days. Each experimental session consisted of the subjects performing the Southampton Hand Assessment Procedure (SHAP) using one of the myoelectric control systems (C1 – i-limb controller, C2 – Postural Controller). We reversed the order of use for each subject to prevent a learning effect between days. Each subject attempted the SHAP twice, once with each controller. The subjects were introduced to the control methodology over a short period (approximately 15 minutes) after which the SHAP was completed. Afterwards, the subjects completed a brief survey to assess qualitatively their opinion on the control methodology. The SHAP times were entered into the SHAP website (http://www.shap.ecs.soton.ac.uk/) which produced the SHAP score (SS, an integer between 0–100 which indicates the level of hand function where ‘0’ is no function and ‘100’ is able-bodied performance) and the functionality profile (an integer between 0–100 for each functional grasp). The SHAP is a time-based assessment method which can be affected by the temporal properties of the prosthetic limb system. In this work, the prosthetic systems were essentially identical between experimental sessions in order to ensure equal prosthetic system speeds across conditions. The SHAP does not elucidate the entire experience of the subject when using the prosthetic limb system and other measures are necessary to assess the entire experience. In this case, a qualitative survey was also deployed (Table 2) using a 5 point Likert scale. Subjects completed the survey after each experimental session and provided verbal comments recorded by the experimenter.

Table 2. –

Qualitative survey provided to subjects following each experimental session.

Item Number Statement Strongly Agree Agree Neutral Disagree Strongly Disagree
1 I was able to learn the PC/i-limb algorithm in less than 15 minutes. ❏ ❏ ❏ ❏ ❏
2 I prefer using the PC/i-limb algorithm over my current method of use. ❏ ❏ ❏ ❏ ❏
3 The ability to control the hand’s closing or opening speed with PC/i-limb was adequate for my use. ❏ ❏ ❏ ❏ ❏
4 The PC/i-limb algorithm is not physically tiresome. ❏ ❏ ❏ ❏ ❏
5 The PC/i-limb algorithm is not mentally tiresome. ❏ ❏ ❏ ❏ ❏
6 I was able to achieve 3 or more grasps easily. ❏ ❏ ❏ ❏ ❏
7 The PC/i-limb algorithm would make every activities more accessible compared to my current method of use ❏ ❏ ❏ ❏ ❏
8 I want to be fitted with the PC/i-limb algorithm today ❏ ❏ ❏ ❏ ❏
Open in a new tab

III. Results

A. Controller Comparison

The i-limb state-based control algorithm (C1) and postural control algorithm (C2) varied between experimental sessions while the rest of the experimental setup was held constant. The prosthetic socket, electrodes, batteries, and prosthetic hands were (essentially) identical between SHAP attempts; we only varied the control strategy. The SS comparison between C1 and C2 produced mixed results between each subject (Figure 3). S1 achieved a greater SS using C2 (63) compared to C1 (58). S1 did not have experience with multi-functional prosthetic hands and was thereby a novice using advanced control methods like C1 and C2. In both cases, S1 was able to learn the control schemes within a short time period (~15 minutes) before completing the SHAP. The experimenter observed that S1 had difficulty performing multiple trigger commands in order to change grasp states when using C1. This difficulty could have caused the lower SS for C1 compared to C2. S2 achieved a higher SS using C1 (81) compared to C2 (68). S2 attributed this difference in performance when using C1 to his extensive experience with C1 while using the i-limb revolution prosthesis prior to experimental session. In general, S2 is an expert user of all types of prosthetic hands and achieved the highest SS using C2 recorded by the authors (compared to all prior subjects 32). The ability of S2 to achieve such a high SS when using a well-practiced control method (C1) may suggest that his SS using C2 would improve after extended use. In fact, S2 was able to perform the SHAP with only a 16% difference between the well-practiced controller (C1) and the untrained postural control system (C2). While this comparison is confounded by the algorithmic differences between C1 and C2, it indicates that a new user of C2 could nearly match the abilities of an experienced user of C1. In general, the SS using C2 of 63 and 68 indicate that approximately 66% of natural hand function is achieved using the prosthetic limb system.

Figure 3. –

Figure 3. –

Open in a new tab

The raw SHAP scores for subjects across the iLimb controller (C1) and postural control algorithm (C2).

B. Prosthetic Hand Comparison

Prior work by the authors studied the postural control algorithm with a modified Bebionic hand27 with four persons with transradial amputation. Since both studies used the SHAP to assess the ability of the subjects, we can make a direct comparison between prosthetic limb systems (Figure 4). The SS when using the i-limb/robo-limb hands was 11 points higher than when using the modified Bebionic hand (a SS of 66 compared to 55, respectively). We attribute this increase in SS to the electromechanical performance of the prosthesis (i.e. – the speed, strength, and kinematics of each hand). In both cases, a glove covered the prosthetic hand to provide a high-friction surface in order to produce more stable grasps of heavy and smooth objects during the SHAP. The functionality profile (Figure 4) details the differences between prosthetic limb systems by functional grasp. The only two grasps with noticeable differences (Tip and Tripod grasps) require direct opposition of thumb and index/middle fingers. The results indicate that the i-limb revolution provided a more functional tip grasp (41 compared to 29, respectively) and tripod grasp (35 compared to 22) however these results may not indicate a functional difference between the two prosthetic hands. We observed that the opposition of the i-limb revolution thumb and index/middle fingers is more stable than with the modified Bebionic hand due to the differing fingertip geometries and hand kinematics. In other words, the ability to form stable and reliable tip/tripod grasps improved the subject’s ability to perform activities of daily living and thereby increase the SS when using the i-limb revolution compared to the modified Bebionic hand.

Figure 4. –

Figure 4. –

Open in a new tab

Comparisons of SHAP scores and functionality profile for Postural Control algorithm with iLimb and Bebionic prosthetic hands.

C. Qualitative Results

Both subjects completed surveys after each experimental session (Table 3). The Likert scale used indicated the extent that the subjects agreed/disagreed with each statement. The results of the survey indicate individual opinions between the two subjects and do not provide trends with which to generalize the results. However, both subjects indicated that when using C2 they were “able to achieve 3 or more grasps easily” and that they “wanted to be fitted with [C2] today”. S2 confirmed in his survey response that he “used [C1] before and liked it.” However, he also confirmed that the i-limb revolution prosthetic hand is “a little weak and little slow.” S1 was not able to perform reliably the required trigger commands when using C1 and therefore strongly disagreed that “the [C1] is not physically tiresome” and that “the [C1] is not mentally tiresome”. S1 described that he “was not happy with how [C1] functioned”. These results confirm many of the observations made by the experimenters. S1’s limited background using advanced myoelectric control systems required additional coaching and practice when learning C1 and C2. However, the lack of triggers when using C2 provided a great benefit to S1. On the other hand, the expert use of C1 by S2 showed that once users learn the trigger commands effectively then the control method is a strong and robust system. Even though S2 achieved a higher SS using C1, he still indicated that he was interested in using C2 in a take-home setting.

Table 3. –

Qualitative survey results for both subjects and both controllers. Green cells indicate agreement, gray cells indicate neutral opinions, and red cells indicate disagreement with the statements provided.

Item Number Statement Postural Control (S1) Postural Control (S2) i-Limb (S1) i-Limb (S2)
1 I was able to learn the PC/i-limb algorithm in less than 15 minutes. Agree Neutral Disagree Strongly Agree
2 I prefer using the PC/i-limb algorithm over my current method of use. Neutral Disagree Strongly Disagree Neutral
3 The ability to control the hand’s closing or opening speed with PC/i-limb was adequate for my use. Neutral Disagree Disagree Strongly Disagree
4 The PC/i-limb algorithm is not physically tiresome. Neutral Neutral Strongly Disagree Agree
5 The PC/i-limb algorithm is not mentally tiresome. Neutral Neutral Strongly Disagree Agree
6 I was able to achieve 3 or more grasps easily. Agree Agree Disagree Strongly Agree
7 The PC/i-limb algorithm would make every activities more accessible compared to my current method of use Neutral Disagree Strongly Disagree Agree
8 I want to be fitted with the PC/i-limb algorithm today Agree Agree Disagree Agree
Open in a new tab

IV. Discussion

A. Postural Control for Clinical Settings

We presented the development of the postural control algorithm throughout a series of publications over the past several years [26,30–32]. While always focused on the clinical implications of the algorithm, this was the first use of the algorithm with all commercially available products and a clinically viable prosthetic socket. A small number of subjects participated in this study due to the expense of creating clinically viable prosthetic limb systems including custom prosthetic sockets, multi-functional prosthetic hands, and lack of time with both Certified Upper Limb Prosthetists as well as subject volunteers. While this study is limited by the small number of subjects (n=2) some critical features of the algorithm and prosthetic system were revealed. The features of the algorithm and prosthesis that were revised in order to build “clinically ready” prosthetic limb systems included 1) a velocity based postural control algorithm to match the clinical standard, 2) a time-based finger positioning method due to lack of motor encoders for position feedback, and 3) a variety of postural control maps for self-selected complexity.

The velocity-based postural control algorithm refers to the deciphering of the myoelectric signal as a velocity (a speed and direction) in the PC domain. This method is in comparison to a position-based controller where the myoelectric signal defines a position in the PC domain. There are several functional differences to these two methods. A lack of EMG activity causes the hand posture to be maintained when using the velocity control while the hand posture returns to hand open when using the position control. In addition, EMG activity is required to open the hand when using the velocity control method while not necessary when using the position control method. We experimented with these two methods previous [31], but only tested in a virtual environment as opposed to a clinically relevant setting. Prosthetists involved in this project directed us to use a velocity-control method in order to mimic the standard “twosite myo” systems that are used today as well as to ensure users were able to “relax” and maintain a grasp. In order to implement the velocity based controller, we needed a method to open the hand in a repeatable fashion (since the hand does not “spring open” like when using a position-based postural controller). This need led us to include a hand open “slice” within the postural control domain. This design change reduced the number of functional grasps available but provided a clinically desirable system.

Many commercially available prosthetic hands are not able to monitor the position of each individual digit. Many prostheses do not include position encoders on the actuators. Prior work on the postural control algorithm used a modified Bebionic hand where position of each digit was monitored using the motor encoders and a custom motor controller board (Sigenics, Inc. Chicago, IL). However, a clinically ready algorithm must be able to command the prosthesis in an open-loop fashion since position feedback is not available in all commercially available multi-functional prosthetic hands. As a result, we redesigned the low-level motor control method to implement the postural controller in an open-loop fashion. First, a calibration is performed each time the hand is powered on to determine the “time to closure” for each digit. Then, a look-up table is built so that each functional grasp is defined by an array of “closure times” for each digit. When the user commands a certain functional grasp, the controller monitors the amount of time and velocity of each digit and thereby calculates the position of the digit. The battery level and silicon glove affect the “closure times” but can be accounted for by power cycling the prosthesis and thereby re-calculate the look-up tables for each functional grasp. This time-based methodology is not as robust as a full closed-loop position control system, but enabled us to implement the postural control algorithm on a commercially available prosthesis.

The use of multiple functional grasps throughout activities of daily living is not well measured among upper limb amputees [33]. The feedback from prosthetists in this project described how many users simply use two or three grasps for nearly all activities. In order to accommodate users with needs for a small or large amount of functional grasps, a customizable postural control domain map was developed. Here the user can choose the number of “slices” available and then populate each slice with the desired grasp. This drag and drop interface ensures customizable maps for each user and the ability to increase the complexity of the algorithm as the user becomes more proficient. In practice, the larger the “slice” the easier (more repeatable) it is for the user to command the desired grasp. As the slices become smaller, then the user must be able to perform repeatable combinations of EMG activity to reach all of the various grasps. This ability was seen to improve with time, but has not been tested over a longitudinal trial.

The postural controller utilizes co-active control signals as opposed to direct control systems with typically “tune out” co-active signals to create two independent signals. The postural control domain includes “slices” which require the co-activity of two electrodes. These co-active regions provide additional control signals to the algorithm thereby increasing the number of grasps/postures. Also, the co-active regions may provide an improved experience for users which cannot create separable or independent myoelectric signals. In this case, a user may choose to use a co-active “slice” for a certain functional grasp since a certain independent EMG signal was too difficult to produce reliably. The customizability of the postural controller for all forms of EMG activity will hopefully provide a clinically robust algorithm for all persons with trans-radial amputation.

B. Comparison to Pattern Recognition

In recent years, multiple commercial entities have released myoelectric pattern recognition products (Coapt LLC and Infinite Biomedical Technologies, LLC) or presented products to be released in the near future (Otto Bock Inc. from MEC proceedings). The field of upper limb prosthetic control studied the mathematical methods used in these algorithms for many years [14], but the functional requirements across all implementations of pattern recognitions algorithms include 1) the requirement of six or more electrodes and 2) regular algorithmic training sessions. These two attributes of pattern recognition systems are major hurdles to the clinical implementation of these systems. Major strides were made on both requirements by commercial entities like Coapt, Infinite Biomedical Technologies, and Otto Bock to alleviate these concerns. The design of low-profile electrodes and compact electronic packages allows more electrodes to be fit into a smaller socket. This is especially significant for short trans-radial and/or trans-humeral amputees where minimal musculature is left in the residual limb. Nonetheless, the cost of additional electrodes may be a hurdle to some users who would otherwise benefit from the technology. Novel algorithmic training methods allow for “on-the-fly” retraining of the pattern recognition algorithm to augment the reliability of the system. Lock et al. [34,35] developed a method, namely “prosthesis guided training”, where the prosthesis moves through a preset sequence of motor functions while the user provides training data to the prosthesis. The progress made by the field of myoelectric control on pattern recognition algorithms have finally allowed the technology to get out of the lab and into the clinic, but these fundamental hurdles of 1) multiple electrode fittings and 2) algorithmic training remain.

In comparison, the postural control algorithm alleviates these concerns while still providing a myoelectric control interface for multi-functional prosthetic hands. The postural controller presented here uses three electrodes (instead of the typical two-electrode setup). Previously, we also implemented a two-site postural controller [32] due to the short residual limb of the subject. In general, the postural controller can be implemented with the standard two-site electrode prosthesis and/or a three-site system thereby alleviating the additional complexity and cost of a multiple electrode pattern recognition fitting. Additionally, minimizing the number of holes through the socket wall that need to be sealed is advantageous especially in the case of suction suspension. The postural controller does not require any algorithmic training whatsoever. The mathematical basis of the postural controller includes vector summation and linear algebra [31] and thereby does not require the training dataset used in pattern recognition systems. The mathematical basis of the algorithm avoids the use of a classifier so that the clinical challenge of algorithmic training can be avoided completely. These attributes of the postural control algorithm will hopefully allow it to be integrated into commercial prosthetic limb systems in the near future and avoid some of the challenges of the currently available pattern recognition systems.

The postural control algorithm is for the control of hands as currently implemented it does not allow for the control of an entire limb (i.e. – hand, wrist, and elbow) while some pattern recognition systems can command all joints of the limb. With that said, the postural control algorithm may be an effective addition to a pattern recognition system where the pattern recognition system commands the elbow and wrist while the postural control algorithm controls the hand. The postural controller also requires accurate site selection in order to acquire three independent EMG signals. Pattern recognition systems do not require specific locations of the electrode array with respect to the limb and therefore reduces the burden on the prosthetist during the fitting. Finally, the postural control algorithm requires the learning of a joystick interface where the EMG activity drives a cursor within a two-dimensional space like a joystick. The PC domain has not physiological relevance and is therefore a seemingly unintuitive method for control hand grasps/postures. Pattern recognition systems could be trained to use physiologically relevant patterns of control signals to control hand grasps/postures. In other words, the intuitiveness of the postural controller is lacking compared to what a pattern recognition system could provide. The subjects did not mention a difference in the cognitive burden of the postural controller as compared to their current system (Table 3), but neither subject is a regular user of a pattern recognition system.

V. Conclusion

We studied the clinical readiness of the postural control algorithm in a laboratory setting with two trans-radial amputees using a commercially available prosthetic limb system. A comparison between the i-limb revolution state-based myoelectric controller and the postural controller was performed using the Southampton Hand Assessment Procedure. The SHAP scores indicate that the postural controller with i-limb revolution provided an average of 66% of hand function compared to an intact limb. This performance was limited by the control interface and the electromechanical limitations of the hand including speed and positional feedback. Future work may seek out commercial prosthetic devices with better electromechanical specifications like the TASKA hand (TASKA Prosthetics, Christchurch New Zealand). The performance using the i-limb revolution state-based controller versus the postural controller was mixed. Various aspects of the postural control algorithm were modified in order to suit the clinical needs of trans-radial amputees. Finally, we provide a discussion of the differences between commercially available pattern recognition algorithms and the postural controller. In the future, the authors plan to provide take-home prostheses with data logging capability to monitor the use of the postural control algorithm in a home setting. We will recruit trans-radial amputees and test in the laboratory before and after the take-home trial to quantify the learning effects of longitudinal use.

Acknowledgments

* This work was supported in part by Career Development Award Number IK1RX00201 from the United States (U.S.) Department of Veterans Affairs Rehabilitation R&D (Rehab RD) Service and the National Institute of Health award (1R44HD090811-01).

Contributor Information

Jacob L. Segil, Rocky Mountain Regional VA Medical Center and the Engineering Plus Program at the University of Colorado Boulder, Boulder CO, 80304.

Rahul Kaliki, Infinite Biomedical Technologies, Baltimore, MD 21202.

Jack Uellendahl, Hanger Clinic, Phoenix, Arizona, 80514.

Richard F. ff. Weir, Rocky Mountain Regional VA Medical Center and the University of Colorado Denver | Anschutz Medical Campus Aurora, CO 80045.

VI. References

  • [1].Biddiss E, Beaton D, Chau T. Consumer design priorities for upper limb prosthetics. Disabil Rehabil Assist Technol. 2007;2(6):346–357. doi: 10.1080/17483100701714733. [DOI] [PubMed] [Google Scholar]
  • [2].Jette AM, Spicer CM, Flaubert JL, eds. The Promise of Assistive Technology to Enhance Activity and Work Participation. Washington, D.C.: National Academies Press; 2017. https://www.nap.edu/catalog/24740. Accessed November 9, 2017. [PubMed] [Google Scholar]
  • [3].Burck JM, Bigelow JD, Harshbarger SD. Revolutionizing prosthetics: Systems engineering challenges and opportunities. Johns Hopkins APL Tech Dig. 2011;30(3):186–197. [Google Scholar]
  • [4].Belter JT, Segil JL, Dollar AM, Weir RF. Mechanical design and performance specifications of anthropomorphic prosthetic hands: A review. J Rehabil Res Dev Wash. 2013;50(5):599–618. [DOI] [PubMed] [Google Scholar]
  • [5].Childress DS, Weir RF. Control of limb prostheses. Atlas Limb Prosthet. 2004; 2:175–198. [Google Scholar]
  • [6].Bongers RM, Kyberd PJ, Bouwsema H, Kenney LPJ, Plettenburg DH, Van der Sluis CK. Bernstein’s Levels of Construction of Movements Applied to Upper Limb Prosthetics. JPO J Prosthet Orthot. 2012;24(2):67–76. doi: 10.1097/JPO.0b013e3182532419. [DOI] [Google Scholar]
  • [7].Roche AD, Rehbaum H, Farina D, Aszmann OC. Prosthetic Myoelectric Control Strategies: A Clinical Perspective. Curr Surg Rep. 2014;2(3):44. doi: 10.1007/s40137-013-0044-8. [DOI] [Google Scholar]
  • [8].Merletti R, Knaflitz M, Luca CJD. Myoelectric manifestations of fatigue in voluntary and electrically elicited contractions. J Appl Physiol. 1990;69(5):1810–1820. [DOI] [PubMed] [Google Scholar]
  • [9].Park E, Meek SG. Fatigue compensation of the electromyographic signal for prosthetic control and force estimation. IEEE Trans Biomed Eng. 1993;40(10):1019–1023. [DOI] [PubMed] [Google Scholar]
  • [10].Castellini C, van der Smagt P. Surface EMG in advanced hand prosthetics. Biol Cybern. 2009;100(1):35–47. doi: 10.1007/s00422-0080278-1. [DOI] [PubMed] [Google Scholar]
  • [11].McHugh S Experiencing morph for the first time. May 2012.
  • [12].Englehart K, Hudgins B. A robust, real-time control scheme for multifunction myoelectric control. IEEE Trans Biomed Eng. 2003;50(7):848–854. doi: 10.1109/TBME.2003.813539. [DOI] [PubMed] [Google Scholar]
  • [13].Parker P, Englehart K, Hudgins B. Myoelectric signal processing for control of powered limb prostheses. J Electromyogr Kinesiol. 2006;16(6):541–548. [DOI] [PubMed] [Google Scholar]
  • [14].Scheme E, Englehart K. Electromyogram pattern recognition for control of powered upper-limb prostheses: state of the art and challenges for clinical use. J Rehabil Res Dev. 2011;48(6):643–659. [DOI] [PubMed] [Google Scholar]
  • [15].Hargrove LJ, Miller LA, Turner K, Kuiken TA. Myoelectric Pattern Recognition Outperforms Direct Control for Transhumeral Amputees with Targeted Muscle Reinnervation: A Randomized Clinical Trial. Sci Rep. 2017;7(1):13840. doi: 10.1038/s41598-017-14386-w. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • [16].Parker PA, Scott RN. Myoelectric control of prostheses. Crit Rev Biomed Eng. 1986;13(4):283–310. [PubMed] [Google Scholar]
  • [17].Li Guanglin, Schultz AE Kuiken TA. Quantifying Pattern Recognition—Based Myoelectric Control of Multifunctional Transradial Prostheses. IEEE Trans Neural Syst Rehabil Eng. 2010;18(2):185–192. doi: 10.1109/TNSRE.2009.2039619. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • [18].Cohen LG, Bandinelli S, Findley TW, Hallett M. MOTOR REORGANIZATION AFTER UPPER LIMB AMPUTATION IN MANA STUDY WITH FOCAL MAGNETIC STIMULATION. Brain. 1991;114(1):615–627. doi: 10.1093/brain/114.1.615. [DOI] [PubMed] [Google Scholar]
  • [19].Röricht S, Meyer B-U, Niehaus L, Brandt SA. Long-term reorganization of motor cortex outputs after arm amputation. Neurology. 1999;53(1):106–106. doi: 10.1212/WNL.53.1.106. [DOI] [PubMed] [Google Scholar]
  • [20].Powell MA, Thakor NV. A Training Strategy for Learning Pattern Recognition Control for Myoelectric Prostheses. J Prosthet Orthot JPO. 2013;25(1):30–41. doi: 10.1097/JPO.0b013e31827af7c1. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • [21].Powell MA, Kaliki RR, Thakor NV. User training for pattern recognition-based myoelectric prostheses: Improving phantom limb movement consistency and distinguishability. IEEE Trans Neural Syst Rehabil Eng. 2014;22(3):522–532. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • [22].Fougner A, Scheme E, Chan ADC, Englehart K, Stavdahl Ø. Resolving the Limb Position Effect in Myoelectric Pattern Recognition. IEEE Trans Neural Syst Rehabil Eng. 2011;19(6):644–651. doi: 10.1109/TNSRE.2011.2163529. [DOI] [PubMed] [Google Scholar]
  • [23].Matrone GC, Cipriani C, Secco EL, Magenes G, Carrozza MC, and others, “Principal components analysis based control of a multi-dof underactuated prosthetic hand,” Journal of neuroengineering and rehabilitation, vol. 7, no. 1, p. 16, 2010. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • [24].Pistohl T, Cipriani C, Jackson A, and Nazarpour K, “Abstract and Proportional Myoelectric Control for Multi-Fingered Hand Prostheses,” Ann Biomed Eng, vol. 41, no. 12, pp. 2687–2698, Dec. 2013. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • [25].Dyson M, Barnes J, and Nazarpour K, “Myoelectric control with abstract decoders,” J. Neural Eng, vol. 15, no. 5, p. 056003, Jul. 2018. [DOI] [PubMed] [Google Scholar]
  • [26].Segil JL and f Weir RF, “Design and Validation of a Morphing Myoelectric Hand Posture Controller Based on Principal Component Analysis of Human Grasping,” IEEE Transactions on Neural Systems and Rehabilitation Engineering, vol. 22, no. 2, pp. 249–257, Mar. 2014. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • [27].Betthauser JL, Hunt CL, Osborn LE, et al. Limb Position Tolerant Pattern Recognition for Myoelectric Prosthesis Control with Adaptive Sparse Representations from Extreme Learning. IEEE Trans Biomed Eng. 2017;PP(99):1–1. doi: 10.1109/TBME.2017.2719400. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • [28].Card SK, English WK, Burr B.Jj. Evaluation of Mouse, RateControlled Isometric Joystick, Step Keys, and Text Keys for Text Selection on a CRT. Ergonomics. 1978;21(8):601–613. doi: 10.1080/00140137808931762. [DOI] [Google Scholar]
  • [29].MacKenzie IS. Fitts’ Law As a Research and Design Tool in Humancomputer Interaction. Hum-Comput Interact. 1992;7(1):91–139. doi: 10.1207/s15327051hci0701_3. [DOI] [Google Scholar]
  • [30].Segil JL, Controzzi M, Weir R. F. ff., and Cipriani C, “Comparative study of state-of-the-art myoelectric controllers for multigrasp prosthetic hands,” J Rehabil Res Dev, vol. 51, no. 9, pp. 1439–1454, 2014. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • [31].Segil JL, Weir RF ff. Novel postural control algorithm for control of multifunctional myoelectric prosthetic hands. J Rehabil Res Dev. 2015;52(4):449–466. doi: 10.1682/JRRD.2014.05.0134. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • [32].Segil JL, Huddle SA, others. Functional assessment of a myoelectric postural controller and multi-functional prosthetic hand by persons with trans-radial limb loss. IEEE Trans Neural Syst Rehabil Eng. 2017;25(6):618–627. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • [33].Zheng JZ, Rosa SDL, Dollar AM. An investigation of grasp type and frequency in daily household and machine shop tasks. In: 2011. IEEE International Conference on Robotics and Automation 2011:41694175. doi: 10.1109/ICRA.2011.5980366. [DOI] [Google Scholar]
  • [34].Lock BA, Simon AM, Stubblefield K, Hargrove LJ. Prosthesis-Guided Training For Practical Use Of Pattern Recognition Control Of Prostheses. 2011. https://dukespace.lib.duke.edu/dspace/handle/10161/4713. Accessed November 14, 2017.
  • [35].Simon AM, Lock BA, Stubblefield KA, Hargrove LJ. Prosthesis-Guided Training Increases Functional Wear Time And Improves Tolerance To Malfunctioning Inputs Of Pattern Recognition–Controlled Prostheses. 2011. https://dukespace.lib.duke.edu/dspace/handle/10161/4725. Accessed November 14, 2017.

RESOURCES