By Megan Boucher-Routhier (University of Ottawa,ON). November 9th, 2022.
In cortical circuits, disinhibited activity yields complex spatiotemporal patterns of fluctuation. One example of this occurs during interictal epileptic brain activity where spiral waves are generated, consisting of complex neural activity that rotates around a center of mass. Here, we investigated the complexity of neural activity in healthy and disinhibited networks as well as deep generative neural networks. Neural activity from acute rodent cortical slices was recorded using a high-density multielectrode array. Cortical disinhibition was induced by perfusing slices with a pro-epileptiform artificial cerebrospinal fluid solution containing a potassium channel blocker 4-Aminopyridine (4-AP), decreased magnesium, and increased potassium. The data collected revealed spiral waves that were characterized by a spatially delimited center of mass, a broad distribution of instantaneous phases across individual electrodes, and a decrease in voltage near the center of mass. Analyses of the participation ratio revealed a broad distribution of eigenvalues whose complexity could not be accounted for by low-rank fluctuations. Subsequently, a deep generative adversarial network (GAN) was trained on the spiral waves extracted from our experimental recordings and captured key features of these waves. By adjusting the input of the GAN, the model generated new samples that deviated in systematic ways from the experimental data, thus allowing the exploration of a broad range of states from healthy to pathologically disinhibited neural networks. Results revealed that the complexity of population activity served as a marker of neural activity along a continuum from healthy to disinhibited brain states. These results open avenues for employing GANs to replicate the dynamics of cortical seizures and accelerate the design of optimal neurostimulation aimed at suppressing pathological brain activity.