Detection of characteristic points of ventricular assist device driving signal, using wavelet decomposition

Pneumatic and hydraulic pressure and flow signals, measured on a working ventricular assist device (VAD) during its test, describe its temporal hydrodynamic conditions. Signals registered as time sample series contain characteristic points or fragments, which reflect consecutive stages of VAD pulsatile work. Because of the nature of signals describing biological objects, they often can be time-varying, transient, non-stationary and affected by multi-source noise. It makes, in some situations, characteristic points of pressure-flow curves unseen in the time domain and automatic detection of these important instants is very difficult or even not possible. We proposed a time-frequency (T-F) analysis approach, where signals are decomposed into adaptive, frequency sub-bands, using a wavelet transform (WT), which is known as a suitable tool for biomedical non-stationary signal analysis. As a result of using WT, the multi-resolution T-F representation is obtained, which is sensitive and can detect both long-term trends and dynamic, sudden changes in the input signal. Our research signal database was created as a result of VAD tests performed for different control parameters on a mock circulatory system, designed and made in our Institute. Results of the proposed automatic detection procedure were presented for three types of WT basis function. We work on the application of our study effects in a control algorithm of testing devices for the determination of the critical control parameters of VAD work conditions.