Revolutionizing Prosthetics is developing a robust hybrid neural interface approach that capitalizes on the strengths of individual signal sources and provides a flexible solution set suitable for a breadth of injuries. Sensory feedback is crucial for effective performance of daily activities, so the fully sensorized limb system supports biofidelic feedback options consistent with the hybrid strategy for closed-loop control.
Several types of recording devices are used to record various biological signals from muscles, peripheral nerves, and the cortex for the purpose of motor decoding. Implanted intramuscular electrodes and surface electromyogram electrodes are used to record muscle activity; implantable peripheral nerve electrode intercept signals, propagating along peripheral nerves; and implantable cortical electrode capture spike and local field potentials, near their origins in the primary motor, premotor, and posterior parietal cortices. Collecting all of these signal modalities provides complementary information as well as a certain level of redundancy that maintains long-term high levels of fidelity and provides modularity. This is especially important because we are addressing tetraplegic patients and the full range of upper-extremity amputees—to include shoulder, transhumeral, transradial, and wrist disarticulations in addition to normal interpatient variations. We have a robust neural interface strategy with a strong cortical focus, and support for noninvasive or minimally invasive integration methods as well.
The Revolutionizing Prosthetics program continues to develop neural decoding algorithms that translate electrical signals received from the body into commands for the limb systems or other devices. These algorithms are all supervisory, requiring a representative data set of the desired commands to “train” the algorithms. The signals used for decoding (e.g., electromyographic action potentials, etc.) have a significant impact on the training data and therefore influence the choice for the best type of algorithm. In addition, the complexity of the desired movement can drive the complexity and size of the training data and thus the complexity of the algorithm. Finally, the algorithms are designed with computational and memory constraints in mind, making linear algorithms the most frequently used.
Control System Architecture for the Modular Prosthetic Limb (Johns Hopkins APL Technical Digest, Volume 30, Issue 3, pp. 217–222, 2011)
Revolutionizing Prosthetics: Neuroscience Framework (Johns Hopkins APL Technical Digest, Volume 30, Issue 3, pp. 223–229, 2011)
Revolutionizing Prosthetics: Devices for Neural Integration (Johns Hopkins APL Technical Digest, Volume 30, Issue 3, pp. 230–239, 2011)