Adversarial AI reveals mechanisms and treatments for disorders of consciousness.

Summary

A generative adversarial AI framework reproduced cross-species neurophysiological signatures of consciousness and coma, retrodicted known responses to stimulation, and generated testable mechanistic predictions. It identified disruption of the basal ganglia indirect pathway and enhanced cortical inhibitory-to-inhibitory coupling in DOC, supported by diffusion MRI (n=51) and human RNA-seq with a rat stroke model, and highlighted subthalamic nucleus high-frequency stimulation as a candidate therapy.

Key Findings

• A generative adversarial AI combining discriminative deep nets (>680,000 10-s samples) with interpretable neural field models simulated conscious and comatose brain activity across species.
• Model predictions identified disruption of the basal ganglia indirect pathway (validated by diffusion MRI in 51 DOC patients) and increased cortical inhibitory-to-inhibitory coupling (supported by RNA-seq in 6 human coma patients and a rat stroke model).
• High-frequency subthalamic nucleus stimulation emerged as a promising intervention, supported by human electrophysiological data.

Clinical Implications

Highlights candidate neural targets (e.g., subthalamic nucleus) and mechanistic biomarkers that could guide neuromodulation trials in DOC and inform perioperative strategies for consciousness monitoring and recovery.

Limitations

• Not an interventional clinical trial; therapeutic predictions require prospective testing.
• Heterogeneity across datasets and species may limit direct clinical generalizability.

Future Directions

Prospective neuromodulation trials targeting AI-identified circuits (e.g., STN) in DOC; development of bedside biomarkers derived from model-inferred mechanisms to stratify patients and monitor response.