Improvement of spike train decoder under spike detection and classification errors using support vector machine.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Decoding kinematic variables from motor cortical neuron spike trains is vital for neural prosthetics.
• Spike detection and sorting are critical but error-prone steps in processing neural signals.
• Existing decoding algorithms need to be robust to these potential errors.

Purpose of the Study:

• To investigate the efficacy of nonlinear mapping, specifically Support Vector Machines (SVMs), for decoding neural signals.
• To compare SVMs against traditional linear filters and multilayer perceptrons in handling erroneous spike trains.
• To assess the feasibility of using robust decoders in neuroprosthetic devices with imperfect preprocessing.

Main Methods:

• Simulated spike trains from primary motor cortical neurons were generated with realistic preferred direction distributions.
• Decoding algorithms employing nonlinear mapping (SVM) and linear filters were tested.
• Performance was evaluated, particularly under conditions of simulated spike train errors.

Main Results:

• Support Vector Machines (SVMs) demonstrated superior performance in decoding kinematic variables compared to linear filters, especially with erroneous spike trains.
• SVMs facilitated easier decoder training than multilayer perceptrons.
• The findings highlight the advantage of nonlinear decoding for handling imperfect spike sorting.

Conclusions:

• Nonlinear decoding algorithms, particularly SVMs, are more advantageous than optimal linear filters for motor cortical neural prostheses.
• SVM-based decoders are robust to spike sorting errors, simplifying neuroprosthetic device development.
• This research paves the way for more reliable brain-computer interfaces despite preprocessing limitations.