Background: Being able to control an upper limb prosthesis by means of the signals recorded from the peripheral nerves is not a trivial task. New generations of neural electrodes are able to record this information but the quality of the signal can make difficult the extraction of the useful information. Several techniques have been adopted both for central and peripheral acquisitions in order to remove the noise and/or enhance the electrical activity generated by the brain or carried by the nerves.
New methods: In this review, common spike detection algorithms have been tested on both real and simulated recordings to verify which is the best choice to be applied in a neuroprosthetics context. In particular, the moving average algorithm (MAA), the non-linear energy operator (NEO) and the wavelet denoising (WD) have been implemented and their performance have been tested by means of the number of the detected real positives (RPs) and false positives (FPs).
Results: MAA outperforms the other techniques because it is capable of detecting a high amount of RPs and, compared to NEO, with a reduced number of FPs.
Comparison with existing methods: MAA needs only the information of the duration of the action potential while the NEO and the WD require the frequency and/or the shape of the action potentials.
Conclusions: NEO and WD are algorithms requiring information about the signal, not a priori known. MAA, then, seems most suitable for online applications.
Keywords: Neural signals; Spike detection; Upper limb prosthetics.
Copyright © 2018 Elsevier B.V. All rights reserved.