• machine learning
• microneurographic recordings
• physiological signals
• vagus nerve

Bioelectronic medicine offers novel therapies for drug-resistant pathologies by delivering electrical stimuli to the autonomic nervous system using implantable devices. Vagus nerve stimulation (VNS) is a promising bioelectronic medicine application, approved by the FDA for the clinical treatment of epilepsy and depression, currently under investigation to treat a wide range of diseases. VNS protocols are currently administered applying in a continuous open-loop fashion pre-set stimulation parameters. Closed-loop protocols employing the recording of physiological signals linked to the state of the patient could lead to the formulation of more effective treatments through the adaptation of stimulation parameters. Direct recordings of vagus nerve activity may serve as a proxy of spontaneous and artificially evoked physiological functions and may be recorded with the same implant already used for stimulation. However, this requires understanding how the neural activity of the fibers in the vagus nerve is related to the subject's physiological state.
This thesis explores the decoding of physiological functions from intraneural recording of the vagus nerves. Microneurographic recordings from the vagus nerve of 18 healthy subjects were inspected and the stability of the relationship between respiratory, cardiovascular and neural variables was studied through correlation analysis. Then, a set of recording intervals were identified were machine learning-based decoding of the physiological functions from the neural signal was attempted. We implemented novel decoding strategies and hyperparameter optimization protocols to attain unprecedented levels of performance in the regression of the respiration signal, across slow and normal breathing periods, and in the regression of the ECG R-peak location.
Since it has been shown how microneurography can be employed as a proxy for the estimation of the performance of implanted intraneural interface, our findings will be likely translatable to clinical VNS implants. Our work is thus an important step in the study of physiological interpretation of the electrical activity flowing through the human vagus nerve, and paves the way for closed-loop VNS protocols for more effective therapeutic interventions in respiratory and cardiovascular diseases.