Comprehensive Summary
Jacob et al.’s study “Brainwide hemodynamics predict EEG neural rhythms across sleep and wakefulness in humans” sought to use machine learning to predict neural rhythms on EEG from hemodynamic data from fMRI. The researchers selected two bands of neural frequencies to examine: alpha waves, which are prevalent during eyes-closed wakeful rest, and delta, which are present mainly in the non-REM stage of sleep. The fMRI data was sectioned into 62 cortical, 14 subcortical, and 8 non-gray matter regions. The model developed by Jacob et al. successfully predicted alpha power above chance in brainwide analysis and when focused on cortical and subcortical fMRI signals. Delta power was also significantly predicted by all of the same categories as the alpha power, but was also able to be predicted by the non-gray matter regions on their own. Regional clustering analysis showed that alpha power could be split into two distinct networks, while delta power did not form an identifiable network. The smaller cluster of alpha-predictive regions included the thalamus, the dorsal striatum, the pallidum, the anterior and posterior cingulate cortices, and the transverse temporal cortex, and functionally seems to modulate the arousal circuit. The larger cluster of alpha-predictive regions relates to visual processing and includes the visual cortex, prefrontal and temporal regions associated with higher-order visual cognition, and three deep-brain regions that are responsive to contents of visual information.
Outcomes and Implications
In the study of sleep and vigilance states, neural rhythms defined by the appearance of distinct oscillatory EEG patterns correspond to well-studied cognitive outcomes, including several forms of memory and attention, along with basic physiological processes like the clearance of brain waste. Neural rhythms are typically monitored with EEG, but scalp EEG suffers from low spatial resolution and is unable to resolve deep brain activity. fMRI may not be a sufficient alternative due to its low temporal resolution, but does allow for subcortical and non-gray matter imaging. Jacob et al.’s predictive model for EEG neural rhythms from fMRi allows for more complete network connectivity patterns, like the two distinct alpha power networks they demonstrate, and additionally provides evidence that fMRI can be used to analyze alpha and delta powers.