MESH: Multimodal Estimators for Sensing Health

项目总结/文摘

Project Summary/Abstract The PI’s goal is to develop an interdisciplinary research program and a foundational algorithmic framework for reliably inferring health states from physiological signals acquired using wearable and portable physiological monitoring devices. Uncovering the health states will unleash an array of applications related to monitoring inflammation, metabolism, fatigue and interoceptive awareness. For instance, it is well known that hormones play an important role in maintaining homeostasis of the body, while cytokines are crucial as mediators of immune response after surgery or infection that disturbs this homeostasis. Adverse external influences such as stress can profoundly alter the hormone or cytokine production in patients, affecting their health and complicating recovery from diseases or surgery. The knowledge of their secretion and modulation in response to major influences such as cardiac surgery, medications, disease, and stress is crucial to the health of patients, more so when more than one of these factors is concurrently present. Thus, there is a compelling but unfulfilled need to quantify hidden health states of inflammation, metabolism, fatigue and interoceptive awareness. The PI’s laboratory seeks to pioneer system-theoretic computational toolsets for understanding the pulsatile signaling underlying the physiological signals (e.g., cytokines, hormones, eye movement) related to different health states and capturing the unobserved temporal dynamics of one’s health states in a biologically plausible manner while considering extensive experimental settings and clinical data. This project will determine the pulsatile physiological signaling from discrete, noisy measurements by performing signal deconvolution to extract the neuronal stimuli underlying their modulation, and will build decoders to quantify internal health states that are indicative of inflammation, metabolism, fatigue and interoceptive awareness using both unlabeled information as well as labels via feedback from patients and clinicians, to help physicians interpret physiological data and inform patient-specific treatment in a holistic manner. The proposed research will use de-identified data both from publicly available datasets and those collected by the PI’s collaborators (e.g., endocrinology, rheumatology, neurosurgery, psychiatry, neuroscience) using wearable or portable devices to perform signal analysis and compare the results against previously published results, known experimental settings, and clinical knowledge to validate the models and provide new insight.