Modeling and analysis of cardioimpedance signals using polynomial models and fuzzy rule-based models.

The determination of characteristic points of bioimpedance curves is crucial to the analysis of bioimpedance signals. Traditional methods need to calculate the derivative of the bioimpedance curve and rely on the help of other simultaneously recorded signals. However, these methods are sensitive to noise while some auxiliary signal may be unrecorded. A novel approach is proposed in this study to automatically determine the positions of characteristic points. The overall development process is realized as a two-phase construct. The first stage of the process involves a fitting of the impedance change curves using nonlinear regression models. Then, a prediction model is trained to predict the positions of characteristic points through a linear combination of the parameters of the nonlinear regression model. Experimental studies constructed on a collection of real-world bioimpedance signals help quantify the performance of the algorithm and gain a deep insight into the superiority of the proposed methodology. It is shown that the proposed method offers a substantial improvement in the prediction accuracy in comparison with the baseline method. Moreover, the proposed characteristic point determination method is robust to noise since its performance is not affected in the presence of noise. • The aim is to automatically determine the positions of characteristic points of bioimpedance signals. • Nonlinear regression models are used to capture the nonlinear and non-stationary characteristics of bioimpedance signals. • The proposed method works well when bioimpedance signals are usually corrupted by noises. • Traditional existing methods are sensitive to noise and rely on other auxiliary signals to work. [ABSTRACT FROM AUTHOR]