Characterization and processing of novel neck photoplethysmography signals for cardiorespiratory monitoring

Abstract

Epilepsy is a neurological disorder causing serious brain seizures that severely affect the patients' quality of life. Sudden unexpected death in epilepsy (SUDEP), for which no evident decease reason is found after post-mortem examination, is a common cause of mortality. The mechanisms leading to SUDEP are uncertain, but, centrally mediated apneic respiratory dysfunction, inducing dangerous hypoxemia, plays a key role. Continuous physiological monitoring appears as the only reliable solution for SUDEP prevention. However, current seizure-detection systems do not show enough sensitivity and present a high number of intolerable false alarms. A wearable system capable of measuring several physiological signals from the same body location, could efficiently overcome these limitations. In this framework, a neck wearable apnea detection device (WADD), sensing airflow through tracheal sounds, was designed. Despite the promising performance, it is still necessary to integrate an oximeter sensor into the system, to measure oxygen saturation in blood (SpO2) from neck photoplethysmography (PPG) signals, and hence, support the apnea detection decision. The neck is a novel PPG measurement site that has not yet been thoroughly explored, due to numerous challenges. This research work aims to characterize neck PPG signals, in order to fully exploit this alternative pulse oximetry location, for precise cardiorespiratory biomarkers monitoring. In this thesis, neck PPG signals were recorded, for the first time in literature, in a series of experiments under different artifacts and respiratory conditions. Morphological and spectral characteristics were analyzed in order to identify potential singularities of the signals. The most common neck PPG artifacts critically corrupting the signal quality, and other breathing states of interest, were thoroughly characterized in terms of the most discriminative features. An algorithm was further developed to differentiate artifacts from clean PPG signals. Both, the proposed characterization and classification model can be useful tools for researchers to denoise neck PPG signals and exploit them in a variety of clinical contexts. In addition to that, it was demonstrated that the neck also offered the possibility, unlike other body parts, to extract the Jugular Venous Pulse (JVP) non-invasively. Overall, the thesis showed how the neck could be an optimum location for multi-modal monitoring in the context of diseases affecting respiration, since it not only allows the sensing of airflow related signals, but also, the breathing frequency component of the PPG appeared more prominent than in the standard finger location. In this context, this property enabled the extraction of relevant features to develop a promising algorithm for apnea detection in near-real time. These findings could be of great importance for SUDEP prevention, facilitating the investigation of the mechanisms and risk factors associated to it, and ultimately reduce epilepsy mortality.