It is difficult to create intuitive methods of controlling prosthetic limbs, often resulting in abandonment. Peripheral nerve interface signals can be used to convert motor intent into usable commands. Extraneural Spatiotemporal Compound Action Potentials Extraction (ESCAPE) is a deep learning-based framework that was demonstrated to be effective at discriminating neural sources in animal models. For clinical application, ESCAPE needs to be validated in humans.
Using cross-sectional immunohistochemistry images of a human median nerve, a finite element model was generated and used to simulate extraneural recordings. ESCAPE was used to classify naturally-evoked compound action potential (nCAPs) based on source location.
Classification accuracy was found to be inversely related to the number of nCAP sources: (3- class: 0.96 +/- 0.03; 10-class: 0.89 +/- 0.07 in low-noise conditions). The performance observed compared favorably to results in animal models. These results demonstrate the promise of ESCAPE for use in human nerves in vivo.