PROJECT SUMMARY/ABSTRACT The growing availability of large functional magnetic resonance imaging (fMRI) datasets has enabled new investigations into functional systems of the human brain. A challenge – but also opportunity – of fMRI arises from the fact that BOLD signal stems from multiple intertwined neural and physiological sources. One major contributor to fMRI signals arises from slow (<0.15 Hz) fluctuations in respiration volume (RV) and heart rate (HR); these systemic physiological fluctuations can account for a substantial proportion of fMRI signals across gray matter, and exhibit spatial patterns that overlap with functional networks. While often treated as a confound, the components of fMRI data linked with systemic physiology may itself present useful information about brain function and physiology, enabling novel investigation of brain vasculature, autonomic function, and brain-body interactions. However, many existing fMRI datasets lack concurrent physiological recordings, and current data-driven techniques do not unambiguously resolve low-frequency physiological signal sources without peripheral cardiac and respiratory recordings for reference. This proposal conducts novel analyses to establish associations between fMRI physiological responses, brain networks, and neurocognitive function. Further, new techniques are proposed for extracting RV and HR time series directly from fMRI data, thereby enriching existing fMRI datasets with missing physiological information. Through analysis of large, public datasets, we will: 1) optimize and validate a deep learning technique for reconstructing physiological time series from resting-state fMRI data alone, which generalizes to participants across the adult lifespan; and 2) relate brain-wide fMRI physiological features to age and phenotypic variation; and 3) probe the value of fMRI physiological responses as early markers of Alzheimer's Disease. We will make all of the resulting signals, models, and code readily available to the community, so that researchers can apply and extend our methods to enhance the value of many existing datasets. Through approaches for resolving neural and physiological sources underlying fMRI signal dynamics, this project also has implications for increasing the precision of fMRI for mapping brain circuits at the level of the individual.