Artifact rejection and cycle detection in immature breathing: Application to the early detection of neonatal sepsis

International audienceThis paper proposes a new framework to obtain quality respiratory variability signals from the raw breathing recorded in neonatal intensive care units (NICUs). It combines three consecutive blocks: an automatic rejection of artifacts, implemented by a logistic regression classifier, a two-step filtering process, and the identification of respiratory cycles, imple-mented by a peak detection algorithm. By means of a gold standard built from a preterm infants database, the performances of the first and third blocks have been evaluated. While the former obtains a 86% of specificity and sensitivity, the latter attains a respective 97%. The interest of our proposal in the clinical domain is illustrated by a promising application to detect promptly and non-invasively the presence of neonatal sepsis in the NICU