Does an artificial neural network model improve the accuracy of reconstructing estimated brachial pressure from finger pressure waves compared to previous methods in healthy persons?
An artificial neural network model can accurately reconstruct estimated brachial pressure from finger pressure waves, outperforming previous linear methods.
In order to correct the wave distortion and pressure decrement (-8.31± 10.29 mmHg) between estimated brachial artery pressure (eBAP) and finger artery pressure (FinAP), we proposed a model based on artificial neural network (ANN). Firstly, we derived 20 morphological properties of FinAP pressure wave and chose 18 strong related of them after correlation analysis with BAP waves. Then, the proposed ANN model was trained with plenty of data which derived from 10 healthy persons in three experimental conditions (SS, BFF and BFC, see TABLE III). Unlike the most linear model, this nonlinear model, which yielded a smaller error between reconstructed brachial pressure and targets (0.16± 1.06 mmHg) than previous Filter (-0.65± 2.75 mmHg) and TFRon (2.06± 1.68 mmHg) methods in total waveform, could predict values in smaller range of error. While superior performance of ANN reflected in the dramatic changes of blood pressure, the error declined to 0.15± 0.88 mmHg in BFF phase and 0.52± 0.88 mmHg in BFC phase in total waveform.
Cai et al. (Tue,) studied this question.