Understanding the neural signatures of consciousness and the mechanisms underlying its disorders, such as coma and unresponsive wakefulness syndrome, remains a critical challenge in neuroscience. In this study, we present a novel computational approach for the in silico discovery of neural correlates of consciousness, the mechanisms driving its disorders, and potential treatment strategies. Inspired by generative adversarial networks, which have driven recent advancements in generative artificial intelligence (AI), we trained deep neural networks to detect consciousness across multiple brain areas and species, including humans. These networks were then integrated with a genetic algorithm to optimize a brain-wide mean-field model of neural electrodynamics. The result is a realistic simulation of conscious brain states and disorders of consciousness (DOC), which not only recapitulates known mechanisms of unconsciousness but also predicts novel causes expected to lead to these conditions. Beyond simulating DOC, our model provides a platform for exploring therapeutic interventions, specifically deep brain stimulation (DBS), which has shown promise in improving levels of awareness in DOC in over five decades of study. We systematically applied simulated DBS to various brain regions at a wide range of frequencies to identify an optimal paradigm for reigniting consciousness in this cohort. Our findings suggest that in addition to previously studied thalamic and pallidal stimulation, high-frequency stimulation of the subthalamic nucleus, a relatively underexplored target in DOC, may hold significant promise for restoring consciousness in this set of disorders.
Read full abstract