Phase-targeted sleep EEG neurofeedback inside the MR scanner drives cerebrospinal fluid flow

66 

Abstract Submission 

Joshua Levitt¹, Leandro Jacob², Laura Lewis³

¹Boston University, Boston, MA, ²Massachusetts Institute of Technology, Cambridge, MA, ³Massacusetts Institute of Technology, Cambridge, MA

Joshua Levitt, Sc. M.  
Boston University
Boston, MA

Leandro Jacob, Ph.D.  
Massachusetts Institute of Technology
Cambridge, MA

Laura Lewis, Ph.D.  
Massacusetts Institute of Technology
Cambridge, MA

Closed-loop neurofeedback methods hold great promise for enhancing the precision and performance of neurostimulation techniques. However, how closed-loop EEG interventions affect brain function is not well understood, due to the challenges of performing MRI imaging during closed-loop EEG interventions. Recent advances in low latency denoising of EEG have made EEG-fMRI neurofeedback more feasible [1, 2]. Here we developed an algorithm that enables low-latency EEG-fMRI neurofeedback and use it to gain insight into the neurobiological effects of closed-loop acoustic stimulation (CLAS). Previous experimentation with CLAS has shown that delivering auditory stimuli in-phase with sleep slow waves (fig. 1a) improves performance on memory tasks, and increases slow wave duration and amplitude [3, 4]. By performing a CLAS experiment inside the scanner, we collected high resolution spatial data to examine the neural basis of this intriguing finding. We focused in particular on cerebrospinal fluid (CSF) and slow waves, which have been shown to be temporally correlated [5]. Sleep contains pulsatile waves of CSF flow linked to waste clearance [5, 6].

8 adults were recruited to participate in an EEG-fMRI nap study, and each completed two 25-minute sleep runs. During each run, they were instructed to press a button with each breath until they fell asleep. A neural network was used to predict upcoming slow wave phase of channel FpZ. When the phase was predicted to fall within a desired range corresponding to the slow-wave peak, the subject randomly received either an audio stimulus (50ms of pink noise; 50% chance) or a sham stimulus (no stimulus; 50% chance). EEG data was collected and preprocessed in real time using LLAMAS [1] to remove scanner artifacts (acquisition code shared at github.com/jalevitt/EEG-LLAMAS/).
MR data was collected with a 3T scanner and a TR of 378ms, calling upon recent advances in fast fMRI [7]. Volumes were positioned with the bottom slice at the entrance to the 4th ventricle, and CSF flow was measured as in Williams et al. [5, 8].
Stimuli delivered while the subject was awake were excluded from analysis, as were stimuli for which the delta power in the previous 10 seconds was below 3uV^2, to remove those not delivered during a slow wave.

We calculated the slow-wave phase at the time of the stimulus and found successful phase-targeting of slow waves (fig. 1 b) and found a significant difference between the stim and sham ERPs (fig. 1c). To assess the effect of the stimulus on oscillatory dynamics, we calculated the mean event-locked power in the spindle band (13-16Hz) and slow wave band (0.4-3Hz) in the time range 0.5-1s . We found that stimulation caused a significant increase in slow wave power and spindle power (fig. 2a-c).
We calculated the stimulus-evoked response of the CSF signal and found a significant increase in CSF flow after stimulation in the stim condition compared to the sham condition (fig. 2d).

We successfully performed EEG-fMRI neurofeedback, enabling us to image the neural consequences of EEG-targeted sensory stimulation in sleep. Furthermore, we show that were able to deliver stimuli in-phase with slow waves using a neural network, which is a novel approach. Our results replicate prior studies that an evoked response and enhanced slow-wave power with CLAS. We also found an increase in spindle power, which supports the hypothesis that spindles act as a 'sensory gate' [9].
We found that phase-targeted stimulation increased CSF flow into the 4th ventricle. Previous research found an increase in CSF flow following slow waves [5], and our results demonstrate this effect can be causally increased using an audio stimulus to enhance slow waves.
This study establishes the feasibility of EEG-fMRI neurofeedback, enabling a wide range of studies to image the effects of EEG-based neuromodulation. Furthermore, our results demonstrate that CLAS can provide a noninvasive way to enhance CSF flow in the human brain.

Non-Invasive Stimulation Methods Other ¹

EEG/MEG Modeling and Analysis

EEG

Multi-Modal Imaging ²

Sleep and Wakefulness

Cerebro Spinal Fluid (CSF)

Electroencephaolography (EEG)

FUNCTIONAL MRI

Machine Learning

Sleep

Other - neurofeedback