UC San Diego

Pathogen infection results in a host inflammation response (cytokines) signaling that can result in systemic inflammatory response syndromes such as sepsis. Cytokines recursively direct the host immune response to infection, playing a key role in immune system configuration, but dysregulated or excessive cytokines contribute to morbidity and mortality associated with sepsis. Low-level inflammatory cytokine concentration change leads to almost immediate increased neuronal firing for Vagus and sympathetic nerve. In this work, we demonstrate that unique measurement methodologies are capable of non-invasively recording peripheral nerve action potentials using optically pumped magnetometers (OPM) and an array of conventional surface electrodes. Both techniques were deployed during a human intravenous Lipopolysaccharides (LPS) injection challenge. The LPS challenge simulated the entry of bacterial compounds into the body, resulting in hyperinflammation. A series of signal processing and spike sorting pipelines were systematically developed to carefully record neural action potentials that entrained to various cytokines produced post-LPS injection. By integrating neural activity measurements and cardiac autonomic measurements, a machine learning method was developed to stratify a patient’s inflammation risk as low, medium, and high. To further refine the reliability of the pipeline, a statistical method based on Wasserstein distance for change point detection was deployed to track the progression of inflammation risk. Furthermore, methods were presented to assess inflammation severity post-LPS injection, using measurements from resting state and autonomic stress challenges. Lastly, a bedside monitoring unit for the continuous measurement and recording of neurological and physiological data for inflammation risk detection was designed and assembled. As a separate arm, this dissertation also studies magnetoencephalography (MEG) sensor-based brain-computer-interface (BCI) for decoding Rock-Paper-Scissor gestures. Unique preprocessing pipelines in tandem with convolutional neural network deep learning models show that non-invasive MEG-based BCI applications hold promise for future BCI development in hand-gesture decoding.