Abstract

Curve fitting for oscillometric waveforms is vital to maximum amplitude algorithm (MAA) in blood pressure measurement. Popular methods in recent years, such as asymmetric Gaussian or Lorentzian functions, perform well when the profile of the oscillometric waveforms (OMW) are close to them. But they will have a relatively large mean square error (MSE) when the oscillometric pulse amplitude envelopes are not so regularly shaped. In this contribution, the artificial neural network (ANN) is implemented instead for the curving fitting. Aided by LabVIEW and MATLAB, its number of neurons in the hidden layer, weight initialization algorithm, training goal and learing algorithm are implemented or carefully considered after some necessary preliminary  work. The experiment  with 48 subjects ranging in age from 18 to 60 years is included in this research. The results show that the back propagation network with 11 neurons in the hidden layer, 0.0025 as training goal and Levenberg-Marquardt learning algorithm is well enough for the curve fitting. ANN with proposed optimum parameters is then compared with the asymmetric Gaussian/Lorentzian functions. After properly adjust the max epoch, ANN can finish computing the envelope in less than a second (3.3 GHz CPU and 4 GB RAM) in all of our experiments like the other two methods while its MSE is still much lower than the other two methods. Their performance in measuring blood pressure is also compared, and ANN shows greater robustness.

This work is licensed under a Creative Commons Attribution 4.0 License.