Pulse oximetry (PO) is a non-invasive medical device used for monitoring of the arterial oxygen saturation (SaO2) and in
particular of haemoglobin oxygenation in blood. Oxygen saturation is commonly used in any setting where the patient
blood oxygen saturation is unstable, including Neonatal Intensive Care Unit (NICU). The main factor affecting PO's
output data is the presence of voluntary or involuntary motion artifacts or imperfect skin-sensor contact. Various
methods have been employed to reject motion artifact but have met with little success.
The aim of the present work is to propose a novel measurement procedure for real-time monitoring and validation of the
oxygen saturation data as measured in standard pulse oxymeter. The procedure should be able to individuate and reject
erroneous saturation data due to incorrect transducer-skin contact or motion artifact. In the case of short sequences of
rejected SpO2 data (time duration< 8s), we report on an algorithm able to substitute the sequence of rejected data with
the "most-probable" (rescued) SpO2 data.
In total we have analyzed 14 patient for a total of 310 hr, 43 min and 15s, equivalent to a total number of samples of
1118595. For our study, we were interested to download heart rate measured with the ECG (HRECG), the heart rate as
measured by the pulse oximeter (HRSAT) and the SpO2 value. In order to remove the erroneous SpO2 values reported in
the rough data in coincidence of motion artifact (top, right), we have implemented a specific algorithm which provides at
the output a new sequence of SpO2 data (validated SpO2 data). With the aim to "rescue" SpO2 value rejected by the previously presented algorithm, we have implemented an algorithm able to provide the "most-probable" SpO2 values in
the case of single rejected values or in the case of short sequences of invalidated data (< 8 s).
From these data it is possible to observe how in the 6.8% of the observation time the SpO2 data measured by the pulse
oximeter are not validated by the use of our method (corresponding to a total time of 16 hr, 8min and 40s). The use of the
proposed algorithm aiming to "rescue" data from short sequences of rejected data (< 8s) allows to increase the validated
data of the 2.5%t(equivalent to 8hr, 40 min and 16s), allowing a percent of usable data of the 95.7%. Once implemented
in clinic, it could be used to identify the period of the day in which the percent of rejected data increase or correlate this
data to clinical procedure in order to intensify clinicians and nurses attention.