Removing Noise from Extracellular Neural Recordings Using Fully Convolutional Denoising Autoencoders.

Area of Science:

• Neuroscience
• Signal Processing
• Artificial Intelligence

Background:

• Extracellular recordings are crucial for neuroscience research but are often contaminated by significant noise.
• Effective denoising is essential for accurate spike sorting and subsequent data analysis.
• Existing denoising methods may not fully address the complexities of neural signal noise.

Purpose of the Study:

• To develop and evaluate an end-to-end deep learning approach for denoising noisy multichannel extracellular recordings.
• To improve the signal-to-noise ratio of neural data for more reliable spike sorting.
• To compare the performance of the proposed deep learning method against established denoising techniques.

Main Methods:

• An end-to-end deep learning model, specifically a Fully Convolutional Denoising Autoencoder, was designed.
• The autoencoder was trained to learn the transformation from noisy multichannel neural signals to clean signals.
• The method was evaluated using simulated noisy neural data.

Main Results:

• The proposed deep learning approach effectively denoises neural signals corrupted by various noise sources.
• Experimental results demonstrate a significant improvement in the quality of denoised neural signals.
• The Fully Convolutional Denoising Autoencoder outperformed traditional wavelet denoising methods in simulated tests.

Conclusions:

• Deep learning, particularly the proposed Fully Convolutional Denoising Autoencoder, offers a powerful solution for denoising extracellular recordings.
• This method enhances the reliability of spike sorting by improving neural signal quality.
• The findings suggest a promising new direction for processing noisy neural data in neuroscience research.