Real-Time Myoelectric Control of Knee and Ankle Motions for Transfemoral Amputees FREE

To the Editor: We recently investigated real-time neural control of artificial arms using targeted muscle reinnervation and pattern recognition algorithms.¹ However, lower limb amputees outnumber upper limb amputees.² There has been increased interest in neurally controlled powered lower limb prosthetics because they can restore activities that require joint power to be generated.^3- 5 We have extended our research to lower limb amputees.

Four transfemoral amputee participants (3 men, 1 woman; mean [SD] age, 50 [9] years; mean [SD] time since amputation, 18 [16.2] years) and 4 intact-limb control participants (2 men, 2 women; mean [SD] age, 29 [2.4] years) were recruited between September 2009 and January 2011 at the Rehabilitation Institute of Chicago. The Northwestern University institutional review board approved the study, and written informed consent was obtained from all participants.

Individuals were seated with bipolar electrodes placed on the skin's surface over 9 muscles: semitendinosus, sartorius, tensor fasciae latae, adductor magnus, gracilis, vastus medialis, rectus femoris, vastus lateralis, and long head of the biceps femoris. Muscle sites were localized based on a combination of normal anatomical locations and palpation⁶ and confirmed by viewing electromyographic (EMG) signals during test contractions (see interactive illustration of lower extremity neuromuscular anatomy). Software¹ instructed participants to complete the following motions: knee flexion and extension, plantar flexion and dorsiflexion, internal and external tibial rotation, internal and external femoral rotation, and relaxation. Twelve seconds of EMG data were collected for each motion from which the computer learned the participants' EMG signal patterns using pattern recognition algorithms.¹ Twelve additional seconds of EMG data were collected for each motion to compute classification accuracy (the percentage of motions correctly predicted by the algorithm). Participants completed virtual environment real-time tests that required them to replicate motions displayed on the computer screen. Trials were successfully completed when the user moved the virtual limb through the complete range of motion that required a minimum of 1 second and were terminated after 15 seconds. There were 2 tests: one completed with a 2−degrees of freedom (DOF) virtual prosthesis and the other a 4-DOF virtual prosthesis (see video of virtual prosthesis testing). Motions were repeated 9 times during 2-DOF tests but only 3 times during 4-DOF tests because more motions were tested.

Performance metrics included classification accuracy, motion completion time, and motion completion percentage.¹ Motion completion time is the time taken from the start of the trial until the virtual limb moved through the complete range of motion. Motion completion percentage is the number of successfully completed motions divided by the total number of trials.

All participants could control both the knee and ankle in the presence of real-time feedback during the 2-DOF test (Figure). All participants also demonstrated 4-DOF control, but with lower performance metrics, particularly for overall motion completion percentage for amputees (Table).

Although neural control of a single DOF at the knee during non–weight-bearing situations has been shown previously,³ this is to our knowledge the first demonstration of neural control of a knee and ankle. Real-time ankle control was unexpected using only EMG signals measured from thigh muscles. These results suggest that targeted muscle reinnervation may not be required to achieve non–weight-bearing control of sagittal plane knee and ankle movements. This is a preliminary study with few participants, and testing was completed in a virtual environment. We are currently modifying powered knee and ankle prostheses^3- 4 to implement our neural control algorithms. Whether these findings will apply when tested on physical prostheses remains to be tested.