Residual Kinetic Vectors for Prosthetic Control

Imagined phantom movements have real physiologic expressions within the residual limb, which we termed residual kinetic activity. In order to exploit the residual functions of transradial amputees we developed a system to register dynamic pressure patterns produced by the residual limb during voluntary commands for motions. This pattern was then converted to a multidimensional vector. Specific requested finger motions were thus detected from the forearm as residual kinetic vectors (RKVs). To decode RKVs, we developed a trainable filter derived from the pseudoinverse of the pressure response matrix. The utility of RKVs was tested on several subjects who expressed both afferent and efferent phantom limb activity. Results showed that amputees using the RKV approach could control at least 3 robotic fingers in near-real time. INTRODUCTION As anthropomorphic robotic hands become more advanced with multiple degrees of freedom [1], their application as a prosthetic device is limited by the ability to measure control signals from amputees. Intensive efforts are underway to enhance prosthetic function by adding more degrees of freedom (DOF) through advanced signal processing of the myoelectric control signals [2,3]. While this approach is promising, it has not yet had a significant clinical impact. As an alternative approach, we have investigated the use of the kinetic, non-electrical, activity of the residual limb as a control signal for prosthesis control. Registering kinetic activity with an array of pressure sensors results in a residual kinetic vector (RKV) that differs fundamentally from the EMG signal. First, the RKV is a vector that dynamically represents the entire limb, not just individual muscle activity. Second, the RKV represents the end result of neuromuscular action, whereas the EMG represents its initiation, and thereby consists of Gaussian noise. Translating Finger Volitions The RKV approach exploits the efferent phantom ability of some below elbow amputees to control extrinsic finger muscles within their residua. This ability represents a phantom efferent phenomenon (PEP) whereby they imagine flexing and extending the metacarpal-phalangeal joints [4]. Their volitions and corresponding muscular activations produce pressure patterns at the residual forearm/socket interface, which are unique for each finger volition. These volitions can be represented as a set of residual kinetic maps. Filtering RKVs A below elbow residuum is a complex mechanical system of bone, muscle, skin, adipose, and scar tissue. Volitional activity is degraded through this system when measured as mechanical activity on the residuum surface by a pressure sensor array. Volition degradation and mechanical coupling could be characterized by a degradation function and its inverse, a restoration function, could discriminate specific movements. Volition degradation is characterized by a system of linear equations g = Hf (1) From “MEC '02 The Next Generation,” Proceedings of the 2002 MyoElectric Controls/Powered Prosthetics Symposium Fredericton, New Brunswick, Canada: August 21–23, 2002. Copyright University of New Brunswick. Distributed under a Creative Commons Attribution-Noncommercial-No Derivative Works 3.0 United States License by UNB and the Institute of Biomedical Engineering, through a partnership with Duke University and the Open Prosthetics Project. where f is a column vector representing intended activity, H is a matrix of the impulse responses, and g is a column vector of measured responses. Approximations of f could be obtained by applying the inverse of the degradation function to the measured responses according to f̂ = Hg. (2) METHODS Human Subjects Subjects were tested following informed written consent, after approval from the Rutgers University Institutional Review Board. An initial screening exam was performed by questionnaire and by direct palpation of the forearm during requested finger motions. Criteria for acceptance into the study were (1) a trans-radial amputation, (2) the presence of afferent and efferent phantom activity, (3) palpable soft tissue movement in the limb. Residual Kinetic Maps Subjects were tested with a thirty-two-element pneumatic sensor array, spaced around the residual limb to detect the mechanical activity of muscle and tendon as the subject performed a volitional movement. The sensors, coupled to pressure transducers, served as inputs to a desktop data acquisition system. Each subject was requested to flex each finger for which he felt capable to move. The data was then interpolated to form a pressure contour map for each volitional movement. Residual KineticVectors The procedure for obtaining the response matrix can be described as follows. Subjects were instructed to perform specific volitions while an 8-element pneumatic sensor array measured mechanical activity on the residuum. A brief phantom finger flexion of maximum intensity was modeled as an impulse to the system, where the component of f corresponding to the volition was assigned a magnitude of 1 and all other components were assigned a value of 0. A response vector was constructed from the maximum amplitude of the signal measured by each sensor channel. Alternatively, multiple repetitions of a particular volition could be performed and RMS values of the responses calculated to provide the response vector components [5]. Repeating the process for other independent volitions that the subject was capable of performing provided unique response vectors. The response vectors associated with specific volitions comprised the columns of the response matrix. Generally, the number of sensors in the array and the number of independent motions were different, so the response matrix was not square (m ≠ n). Therefore, H is the pseudoinverse of H, calculated by the singular value decomposition (SVD) of H [6]. Determination of the weights for the restoration function, or pseudoinverse (PI) filter, occured during an initial training procedure, which was the most computationally demanding step of the process. In contrast, pressure (input) vector filtering can be done in nearly real-time, with control (output) vectors updated before the next acquisition cycle. RESULTS Maximum pressure changes in the socket were approximately 4 kPa. Isobaric contour lines were evenly spaced from the maximum pressure for each volition. Figure 1 shows the residual kinetic maps for subject 1. Each map represents one volitional command over which the subject perceived control. Subject 1 believed he had control over three phantom fingers; therefore we did not attempt to find additional flexion commands. Subject two had different, but equally unique residual kinetic maps for four fingers. From “MEC '02 The Next Generation,” Proceedings of the 2002 MyoElectric Controls/Powered Prosthetics Symposium Fredericton, New Brunswick, Canada: August 21–23, 2002. Copyright University of New Brunswick. Distributed under a Creative Commons Attribution-Noncommercial-No Derivative Works 3.0 United States License by UNB and the Institute of Biomedical Engineering, through a partnership with Duke University and the Open Prosthetics Project. Figure 1. Residual Kinetic Maps of Requested Finger Movements, Subject 1. These pressure contour maps of the residual limb demonstrate the uniqueness of patterns for different requested movements. The abscissa is the distance from the distal end of the residual limb in cm. The ordinate is the normalized circumference in the transverse plane, with 0 being the most lateral point, moving along the anterior surface.