A Sparse Approximation Based Algorithm to Detect Aortic Valve Opening from HVAD Waveforms Acquired via Monitor Snapshot

Abstract

Purpose The HeartWare HVAD monitor displays a real-time waveform of the flow provided by the LVAD. However, there are no commercially available methods to obtain this flow waveform data. We describe a novel algorithm that leverages sparse approximation with Gaussian dictionary to estimate aortic valve opening status from HVAD flow waveform data acquired via monitor snapshot. Methods Stable adult patients receiving MCS therapy with an HeartWare HVAD were included. Aortic valve opening status was determined during routine echocardiography (“Open”\u202f=\u202fopens in ≥8/10 beats vs “Closed”\u202f=\u202fopens in ≤2/10 beats) and a photo of the HVAD monitor was taken with a smartphone at the same time. The flow waveform graph was digitized and interpolated using low frequency Fourier components for the purpose of low pass filtering and standardization of sampling frequency (Figure 1). Sparse approximations (SA) with Gaussian dictionary were generated for each signal using orthogonal matching pursuit (OMP). Waveforms with a closed Ao valve are better aligned with SA (Figure 2: 2a, 2b). Therefore, residual sum of squares (RSS) -which is a marker of deviation from SA- was used to estimate Ao valve status. Results 54 waveforms were acquired. Area under the ROC curve for RSS’ capacity to estimate the status (open/closed) of a single beat was .766 (± .021) (p: .000). The SA-based algorithm had 83.33% overall accuracy for detecting aortic valve opening (sensitivity: 86.67%, specificity: 79.17%) from a 10 sec. signal (Figure 2). Conclusion We describe a method for reliable data acquisition and estimation of aortic valve opening from a HVAD monitor snapshot.