Pulse oximetry is an innovative technique for continuous noninvasive monitoring of arterial blood oxygen saturation value. It has proved to be very useful for many applications such as clinical management of critically ill infants with cardiac and pulmonary diseases and patient monitoring during surgery. Signal processing of the plethysmographic fingerpulse signal for blood oxygen determination is typically completed in the time domain utilizing moving average and peak detection technique. However, it suffers from false detection and errors due to motion artifacts. In this thesis, we present an improved method by computing the scaling factor between the red and infrared ac signal. Experimental results are provided to illustrate the significant improvement....[ Read more ]

Pulse oximetry is an innovative technique for continuous noninvasive monitoring of arterial blood oxygen saturation value. It has proved to be very useful for many applications such as clinical management of critically ill infants with cardiac and pulmonary diseases and patient monitoring during surgery. Signal processing of the plethysmographic fingerpulse signal for blood oxygen determination is typically completed in the time domain utilizing moving average and peak detection technique. However, it suffers from false detection and errors due to motion artifacts. In this thesis, we present an improved method by computing the scaling factor between the red and infrared ac signal. Experimental results are provided to illustrate the significant improvement.

In addition to the studies of blood oxygen level, the problem of cardiac cycle detection and feature extraction from fingerpulse plethysmographic signal is considered. Previous studies have indicated that only rough estimates of the shape of the cardiac cycle together with amplitudes and positions can be obtained by the technique of peak detection and direct segment extraction. To produce a more precise result, the signal is modeled as the convolutional sum of a sequence of impulses with the cardiac cycle, corrupted by noise and offset level. Parameters including the impulse positions, amplitudes and the shape of the cardiac cycle are estimated by a new algorithm using alternating projection which minimizes the mean squared error with regularization constraints. Our technique requires only a rough determination of the pulse positions as an initial estimate and can be applied to any other biosignals of similar types. Experimental results are presented to demonstrate the good reconstruction of the input signals.