DeepPhysioRecon: Tracing peripheral physiology in low frequency fMRI dynamics

Abstract Many studies of the human brain using functional magnetic resonance imaging (fMRI) lack physiological measurements, which substantially impacts the interpretation and richness of fMRI studies. Natural fluctuations in autonomic physiology, such as breathing and heart rate, provide windows into critical functions including cognition, emotion, and health, and can heavily influence fMRI signals. Here, we developed DeepPhysioRecon, a Long-Short-Term-Memory (LSTM)-based network that decodes continuous variations in respiration amplitude and heart rate directly from whole-brain fMRI dynamics. Through systematic evaluations, we investigate the generalizability of this approach across datasets and experimental conditions. We also demonstrate the importance of including these measures in fMRI analyses. This work highlights the importance of studying brain-body interactions, proposes a tool that may enhance the efficacy of fMRI as a biomarker, and provides widely applicable open-source software.