| Abstract | Artifacts in clinical intensive care monitoring lead to false alarms and complicate data analysis. They must be
identified and processed to obtain true information. In this paper, we present a method for detecting artifacts
in heart-rate (HR) and mean blood-pressure (BP) data from a physiological monitoring system used in
preterm infants. The method uses three different types of artifact detectors: limit-based detectors,
deviation-based detectors, and correlation-based detectors. Each identifies artifacts in the monitoring data
from a different perspective. By integrating the individual detectors, we develop a parametric artifact
detector, called CVDetector. The CVDetector is parametric because its performance depends on the
specific values for the parameters in its component detectors. In a huge space of CVDetector instances, we
have successfully discovered an optimal CVDetector instance, denoted by CVDetector*. The sensitivity
and specificity of CVDetector* for HR artifacts is 94.8% (SD=7.6%) and 90.6% (SD=6.9%),
respectively. The sensitivity and specificity of CVDetector* for BP artifacts is 94.2% (SD=5.3%) and
80.0%(SD=12.4%), respectively. |