Sums of Spike Waveform Features for Motor Decoding.

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Spike sorting is crucial for decoding motor intentions from neural recordings.
• Recent methods explore decoding without spike sorting, using waveform features.
• One approach uses waveform feature moments, showing comparable accuracy to spike sorting.

Purpose of the Study:

• To validate and improve decoding approaches that omit spike sorting.
• To investigate using sums of exponentiated waveform features as an alternative to moments.
• To compare decoding accuracy using sorted vs. unsorted spike counts, controlling for 'hash'.

Main Methods:

• Offline analysis of cortical recordings from two Rhesus monkeys.
• Implementation of decoding models using waveform feature moments.
• Development and testing of a novel approach using sums of exponentiated waveform features.
• Comparison of decoding accuracy using sorted spike counts, unsorted threshold crossings, and the new sums-of-features method.

Main Results:

• Using sums of waveform features, particularly spike amplitude, significantly improved hand movement reconstruction accuracy over feature moments.
• The sums-of-amplitude approach outperformed traditional spike sorting and unsorted threshold crossings in offline decoding accuracy.
• When controlling for 'hash' (unmatched waveforms), spike sorting demonstrated superior decoding accuracy compared to unsorted counts.

Conclusions:

• The sums-of-features framework shows potential as an alternative to spike sorting and unsorted threshold crossings for neural decoding.
• Spike amplitude sums offer a computationally inexpensive and effective method for decoding motor intentions.
• Careful consideration of 'hash' is necessary when evaluating the benefits of spike sorting versus unsorted methods.