An Accurate and Robust Method for Spike Sorting Based on Convolutional Neural Networks.

Area of Science:

• Neuroscience
• Biomedical Signal Processing
• Computational Neuroscience

Background:

• Spike sorting is essential for analyzing neural activity from extracellular recordings.
• Accurate separation of neuronal signals is critical for understanding brain function.
• Existing methods face challenges with noisy and overlapping neural signals.

Purpose of the Study:

• To introduce a novel deep learning approach for accurate and robust spike sorting.
• To evaluate the performance of the proposed method on simulated and experimental neural data.
• To demonstrate the advantages of 1D-CNNs for complex spike sorting tasks.

Main Methods:

• Development of a deep learning model utilizing one-dimensional convolutional neural networks (1D-CNNs).
• Validation using simulated extracellular recordings with varying noise levels and spike overlaps.
• Testing on experimental data from a macaque monkey's primary visual cortex.

Main Results:

• Clustering accuracy exceeded 99% on most simulated datasets, even with high noise and spike overlap.
• The 1D-CNN method significantly outperformed the state-of-the-art WMsorting and a multilayer perceptron (MLP) model.
• Successful isolation of spikes from two to five distinct neurons using minimal manually labeled data from experimental recordings.

Conclusions:

• The proposed 1D-CNN deep learning method offers high accuracy and robustness for spike sorting.
• This approach shows significant potential for tackling challenging neuroscientific data analysis, including overlapped spikes and multichannel recordings.
• The method provides a foundation for advanced applications in neural signal processing and brain research.