An improved method for automated breath detection using pharyngeal pressure signals

Introduction: Breath detection in sleep and breathing research is typically a labour intensive task that requires accurate identification of the onset of inspiration and expiration from respiratory signals. Our group previously developed an algorithm to automate this process using the airflow signal and epiglottic pressure signal during periods of restricted airflow and apnoea (Nguyen et al, PLoS One, 2017). However, 11% of breathing efforts were missed with the original algorithm when using the epiglottic pressure signal. This study is an extension of this algorithm development work with the goal of improving accuracy when performing automated breath detection using only the epiglottic pressure signal. Method: Data from previous sleep and breathing research studies in which airflow was measured with a pneumotachograph and nasal mask and negative pharyngeal pressure swings were derived using an epiglottic pressure catheter were analysed. Breath parameters were extracted and then compared between the original and new algorithm. Comparisons were also made using the traditional breath detection approach (zero crossing of the flow signal). Results: The new algorithm improved the speed (>10X reduction in processing speed) and robustness against noise and artefacts, particularly in algorithmically challenging sections pre‐ and post‐ arousals in flow limitation events, reducing the amount of manual verification needed. Discussion: Given that the number of breaths during a full overnight sleep study is substantial, visual analysis of individual breaths for the entire night and for large datasets is not feasible. This new algorithm to automatically detect respiratory effort using the epiglottic pressure signal independent of the flow signal has improved accuracy compared to previous approaches with substantially reduced analytical workload. This tool may be useful for automated approaches for sleep apnoea phenotyping and respiratory physiology research.