Computational approaches to decode grasping force and velocity level in upper-limb amputee from intraneural peripheral signals.

. Recent results have shown the potentials of neural interfaces to provide sensory feedback to subjects with limb amputation increasing prosthesis usability. However, their advantages for decoding motor control signals over current methods based on electromyography (EMG) are still debated. In this study we compared a standard EMG-based method with approaches that use peripheral intraneural data to infer distinct levels of grasping force and velocity in a trans-radial amputee.. Surface EMG (three channels) and intraneural signals (collected with transverse intrafascicular multichannel electrodes, TIMEs, 56 channels) were simultaneously recorded during the amputee's intended grasping movements. We sorted single unit activity (SUA) from each neural signal and then we identified the most informative units. EMG envelopes were extracted from the recorded EMG signals. A reference support vector machine (SVM) classifier was used to map EMG envelopes into desired force and velocity levels. Two decoding approaches using SUA were then tested and compared to the EMG-based reference classifier: (a) SVM classification of firing rates into desired force and velocity levels; (b) reconstruction of covariates (the grasp cue level or EMG envelopes) from neural data and use of covariates for classification into desired force and velocity levels.Using EMG envelopes as reconstructed covariates from SUA yielded significantly better results than the other approaches tested, with performance similar to that of the EMG-based reference classifier, and stable over three different recording days. Of the two reconstruction algorithms used in this approach, a linear Kalman filter and a nonlinear point process adaptive filter, the nonlinear filter gave better results.This study presented a new effective approach for decoding grasping force and velocity from peripheral intraneural signals in a trans-radial amputee, which relies on using SUA to reconstruct EMG envelopes. Being dependent on EMG recordings only for the training phase, this approach can fully exploit the advantages of implanted neural interfaces and potentially overcome, in the medium to long term, current state-of-the-art methods. (Clinical trial's registration number: NCT02848846).