Oscillometric Blood Pressure Estimation Using Principal Component Analysis and Neural Networks Mohamad Forouzanfar, Hilmi R. Dajani, Voicu Z. Groza, Miodrag Bolic School of Information Technology & Engineering (SITE) University of Ottawa Ottawa, ON, Canada {mforo040, hdajani, groza, mbolic}@site.uottawa.ca Sreeraman Rajan Defence Research and Development Canada (DRDC) Ottawa, ON, Canada sreeraman.rajan@drdc-rddc.gc.ca Abstract—Estimation of systolic and diastolic pressures from the oscillometric waveform is a challenging task in noninvasive electronic blood pressure (BP) monitoring devices. Since the conventional oscillometric algorithms cannot model and extract the complex and nonlinear relationship that may exist between BP and oscillometric waveform, artificial neural networks (NNs) have been proposed as a possible alternative. However, the research on this topic has been limited to some simple architectures that directly estimate the BP from raw oscillation amplitudes (OAs). In this paper, we propose principal component analysis as a preprocessing step to decorrelate the OAs and extract the most effective components. Two architectures of NNs, namely, feed-forward and cascade-forward, are employed to estimate the BP using the preprocessed OAs. The networks are trained using the gradient descent with momentum and adaptive learning rate backpropagation algorithm and tested on a dataset of 85 BP waveforms. The performance is then compared with that of the conventional maximum amplitude algorithm and already published NN-based methods. It is found that the proposed networks achieve lower values of mean absolute error and standard deviation of error in estimation of BP compared with the other studied methods. Keywords-blood pressure; oscillometric waveforms; principal component analysis; feed-forward neural network; cascade-forward neural network; estimation I. INTRODUCTION Blood pressure (BP) is one of the vital signs, which along with body temperature, heart rate and respiratory rate carry significant information about physiological state. Hence, it is very important to develop robust methods to accurately measure BP. The most accurate, but not always preferred, method for measuring BP is the invasive arterial measurement. However, due to the difficulty and risk associated with invasive methods, the auscultatory technique is widely used instead. In the auscultatory method, a cuff is placed around the arm at the same height as the heart, and is inflated until the artery is completely occluded. Then, the cuff is slowly deflated by an examiner while listening to the Korotkoff sounds with a stethoscope, in order to identify the diastolic and systolic pressures [1]. Because the traditional auscultatory technique is difficult to automate, electronic BP monitoring devices have been developed based on the so-called “oscillometric” method which estimates the BP from the recorded pressure waveforms [2-4]. The oscillometric method is similar to the auscultatory technique but with a pressure sensor to record the pressure oscillations within the cuff, instead of listening to Korotkoff sounds with sthethoscope. As the occluding cuff is slowly released, pulsatile oscillations appear on the sensed cuff pressure waveform. The amplitude of these oscillations increases to a peak and then decreases with further deflation. Fig. 1 shows a sample cuff pressure waveform along with the extracted oscillations. It is generally accepted that the oscillation amplitudes (OAs) embedded in the cuff pressure carry most of the BP information [5]. After recording these oscillations, analysis algorithms are employed to estimate the diastolic, mean and systolic pressures from these oscillations. The most common oscillometric algorithm is the maximum amplitude algorithm (MAA). The MAA approximates the mean BP as the cuff pressure at which the maximum oscillation occurs and then linearly relates the systolic and diastolic pressures to the mean pressure using empirically derived ratios. These ratios serve to determine the time points at which the cuff pressure coincides with the systolic and diastolic pressures, respectively [3]. In [4], it has been shown that the mean BP may be estimated accurately by MAA. However, due to the sensitivity of the method to variations in BP waveform, Figure 1. Sample cuff pressure waveform (black curve) along with the extracted oscillations (blue curve). The oscillations are magnified for better visualization. The final published version of this paper is available at: http://ieeexplore.ieee.org/xpl/login.jsp?tp=&arnumber=5444353&url=http%3A%2F%2Fieeexplore.ieee.org% 2Fxpls%2Fabs_all.jsp%3Farnumber%3D5444353 Please cite this article as follows: M. Forouzanfar, H.R. Dajani, V.Z. Groza, M. Bolic, and S. Rajan, “Oscillometric blood pressure estimation using principal component analysis and neural networks,” IEEE Toronto Int. Conf. – Science and Technology for Humanity (TIC-STH 2009), Toronto, Canada, Sep. 2009, pp. 981-6.