Extracting spatial-temporal coherent patterns in large-scale neural recordings using dynamic mode decomposition.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Data Science

Background:

• Large-scale neural recordings generate big data with complex spatio-temporal patterns.
• Existing computational methods often analyze spatial or temporal dynamics separately, limiting comprehensive understanding.
• There is a need for methods that can analyze both space and time concurrently in neural data.

Purpose of the Study:

• To adapt Dynamic Mode Decomposition (DMD), a fluid physics algorithm, for analyzing large-scale neural recordings.
• To develop a novel computational approach for visualizing and understanding neural activity.
• To extract meaningful patterns from high-dimensional neural datasets.

Main Methods:

• Adapted Dynamic Mode Decomposition (DMD) for neural data analysis.
• DMD decomposes high-dimensional data into coupled spatial-temporal modes.
• Validated DMD on human motor task recordings and combined it with clustering for sleep spindle network analysis.

Main Results:

• Successfully applied DMD to sub-dural electrode array recordings from human subjects.
• Identified distinct sleep spindle networks characterized by specific cortical distribution, frequency, and duration.
• Demonstrated DMD's robustness to noise and subsampling, and its scalability to large datasets.

Conclusions:

• DMD provides a powerful method for analyzing large-scale neural recordings by integrating spatial and temporal information.
• The approach combines strengths of Principal Component Analysis (PCA) and spectral analysis.
• This technique is well-suited for uncovering complex neural dynamics and network structures.