Rapid-eye-movement sleep (REMS) or paradoxical sleep is associated with intense neuronal activity, fluctuations in autonomic control, body paralysis and brain-wide hyperemia. The mechanisms and functions of these energy-demanding patterns remain elusive and a global picture of brain activation during REMS is currently missing. In the present work, we performed functional ultrasound imaging on rats over multiple coronal and sagittal brain sections during hundreds of spontaneous REMS episodes to provide the spatiotemporal dynamics of vascular activity in 259 brain regions spanning more than 2/3 of the total brain volume. We first demonstrate a dissociation between basal/midbrain and cortical structures, the first ones sustaining tonic activation during REMS while the others are activated in phasic bouts. Second, we isolated the vascular compartment in our recordings and identified arteries in the anterior part of the brain as strongly involved in the blood supply during REMS episodes. Finally, we report a peculiar activation pattern in the posterior amygdala, which is strikingly disconnected from the rest of the brain during most REMS episodes. This last finding suggests that the amygdala undergoes specific processing during REMS and may be linked to the regulation of emotions and the creation of dream content during this very state.

Functional Ultrasound (fUS) provides spatial and temporal frames of the vascular activity in the brain with high resolution and sensitivity in behaving animals. The large amount of resulting data is underused at present due to the lack of appropriate tools to visualize and interpret such signals. Here we show that neural networks can be trained to leverage the richness of information available in fUS datasets to reliably determine behavior, even from a single fUS 2D image after appropriate training. We illustrate the potential of this method with two examples: determining if a rat is moving or static and decoding the animal’s sleep/wake state in a neutral environment. We further demonstrate that our method can be transferred to new recordings, possibly in other animals, without additional training, thereby paving the way for real-time decoding of brain activity based on fUS data. Finally, the learned weights of the network in the latent space were analyzed to extract the relative importance of input data to classify behavior, making this a powerful tool for neuroscientific research.