Hidden-Markov Factor analysis as a spatiotemporal model for electrocorticography.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Signal Processing

Background:

• Electrocorticography (ECoG) signals recorded with high-channel-count arrays offer rich neural data.
• Extracting meaningful low-dimensional neural trajectories from high-dimensional ECoG data is challenging.
• Existing methods like Gaussian-Process Factor Analysis (GPFA) integrate temporal smoothing with dimensionality reduction.

Purpose of the Study:

• To introduce a novel approach, Hidden-Markov Factor Analysis (HMFA), for analyzing high-channel ECoG signals.
• To model the relationship between high-dimensional ECoG neural space and a low-dimensional latent space.
• To enable better understanding of neural correlates of perception and behavior.

Main Methods:

• Developed Hidden-Markov Factor Analysis (HMFA), a probabilistic model linking factor analyzers with a hidden Markov model.
• Focused analysis on the 76-100Hz high gamma sub-band of ECoG signals.
• Applied HMFA to ECoG recordings from 2 subjects performing a button-press task.

Main Results:

• HMFA quantifies ECoG neural space and performs dimensionality reduction in a common probabilistic space.
• A goodness-of-fit metric showed HMFA outperformed GPFA and other spatiotemporal modeling techniques.
• HMFA's performance was superior in predicting electrode activity based on signals from other electrodes.

Conclusions:

• HMFA provides a powerful new tool for extracting low-dimensional neural trajectories from high-channel ECoG data.
• The method effectively models the complex spatiotemporal dynamics of neural signals.
• HMFA offers a promising approach for relating neural activity to perception and behavior.