Classifying neonatal spells using real-time temporal analysis of physiological data streams: Algorithm development

Abstract

Neonatal spells are cardiorespiratory events that occur in newborn infants with variable combinations of cessation of breathing, decrease in blood oxygen saturation and decrease in heart rate. A system using real-time temporal analysis of physiological data streams to accurately detect pauses in breathing together with changes in heart rate and blood oxygen saturation is described. The system uses a multidimensional online health analytics environment that supports the acquisition, transmission and real-time processing of high volume, high rate data. A family of algorithms has been developed using IBM Infosphere Streams, a scalable middleware component for analysing multiple streams of data in real-time. Respiratory pauses are identified by accurately detecting breaths and by calculating the time interval since the last breath. Changes in heart rate and blood oxygen saturation are identified by both threshold breaches and the detection of relative change. The algorithms detect relative change by assessing a sliding normal baseline and generating alerts when values fall out of range. The output of these algorithms has been shown to detect clinically significant relative changes in both heart rate and blood oxygen saturation in a single use case study. The specificity of the algorithm is 98.5%; the sensitivity is 100%. Future research will focus on the application of these algorithms for the assessment and classification of neonatal spells.

A review of data mining using big data in health informatics

“…Thommandram et al [18] use data streams with a different goal attempting to detect and eventually classify neonatal cardiorespiratory spells (a condition that can be greatly helped by being detected and classified in real time). A cardiorespiratory spell is classified as some combination of a pause in breathing, drop in blood oxygen saturation, and a decrease in heart rate.…”

“…Also, data from the other studies presented previously could be used to expand the size of the starting pool (such as MRI data seen in [10], using the relative baseline method from [18], or even message board data from [19] and [20]). Another item to note is that throughout these studies only one feature selection technique was tested with minimal overlap; therefore, future work should look to test multiple feature selection techniques in order to find which one works the best with medical data.…”

“…For the "Real-Time Predictions Using Data Streams" subsection (covered by the research efforts [16,18], and [17]), we see that this area of research is relatively new due to technology only very recently being able to handle the high-throughput processing necessary from the Big Velocity of data streams. As such, research is lacking testing and validation of developed methods as seen for Zhang et al [17] and Thommandram et al [18].…”

“…As such, research is lacking testing and validation of developed methods as seen for Zhang et al [17] and Thommandram et al [18]. As sensors are not perfect (creating missing or erroneous data during a given time period), especially when being used for real-time analysis, future work will need to focus on developing and testing methods that can handle such data in the most reliable and efficient way.…”

Applications of Real-Time Big Data Analytics

“…Zhang et al [12] develop a clinical support system to facilitate real-time prognosis and diagnosis as quick as possible. Thommandram et al [9] designed a system called Artemis to detect cardiorespiratory spell in real-time.…”