A deep generative adversarial network capturing complex spiral waves in disinhibited circuits of the cerebral cortex

Area of Science:

• Neuroscience
• Computational Neuroscience
• Systems Neuroscience

Background:

• Disinhibited brain activity in the cerebral cortex manifests as propagating waves, often in a spiral pattern around a center of mass.
• Spiral waves were hypothesized to be less complex than healthy neural activity due to their stereotypical nature and broad co-fluctuations.

Purpose of the Study:

• To investigate the complexity of spiral waves in disinhibited cortical networks.
• To develop computational models for simulating and analyzing neural activity patterns.

Main Methods:

• Dense multi-electrode recordings of cortical networks induced into a disinhibited state.
• Analysis of neural activity snapshots using participation ratio and eigenspectrum analysis.
• Training a deep generative adversarial network (GAN) to replicate and explore spiral wave dynamics.

Main Results:

• Contrary to hypothesis, spiral waves exhibited increased complexity compared to baseline activity.
• The GAN successfully generated synthetic spiral waves mimicking experimental data features.
• The model allowed exploration of a spectrum of neural network states from healthy to disinhibited.

Conclusions:

• Neural population activity complexity is a marker for brain states, from healthy to disinhibited.
• The developed GAN provides a tool for studying seizure dynamics and designing neurostimulation therapies.