PROJECT SUMMARY
The primary objective of this research project is to create a new approach to identifying the state-of-health
of implanted ventricular assist devices (VAD) using nothing more than a smart electronic stethoscope and a
single lead ECG. VADs are pumps permanently implanted into patients with poor left ventricular function. They
run at a set speed with almost no internal automatic speed adjustments. For this reason, it is a complex problem
for healthcare providers to diagnose VAD patients who present with symptoms that may or may not be due to
reduced pump performance. The current state-of-the-art requires very highly trained healthcare providers with
access to expensive clinical equipment and sometimes requires invasive or partially invasive procedures just to
determine if the symptoms might be due to the VAD device, let alone what the specific problem is.
We hypothesize that through the use of advanced signal processing and machine learning
techniques, we can classify a VAD patient’s health status as it pertains to the VAD’s mechanical
performance and hemodynamic flow into normal and dysfunctional states.
To test this hypothesis, we will modify an existing hemodynamic simulator that creates not only time-varying
pressure to the VAD input and output, but also has heart valves that create an acoustic signature similar to human
valves, and can be underfilled to the point of ventricular collapse (all conditions similar to those commonly seen
in a VAD-rich clinical environment). We will specifically address most VAD failure modes: (1) underfilling of the
VAD inflow relative to the pump speed, (2) excessive afterload due to either systemic hypertension or the
presence of occlusions within the VAD flow circuit, (3) presence of significant valvular disease (aortic
regurgitation, mitral regurgitation, and mitral stenosis), and (4) the effect of heart rate and rhythm on VAD
performance. Various severity levels of these conditions will be seeded in the simulator, and the audio signature
of the VAD pump and associated native heart sounds will be recorded. The digital heart sound measurements
will be processed using novel algorithms to be developed in this protocol to understand the root physical cause,
and compared to the VAD function/dysfunction mode being tested. As the project proceeds actual patient data,
collected by the research team in 2018 in a non-invasive manner (stethoscope, ECG, and ultrasound), will be
compared versus the simulator results under the same conditions.
If the proposed research is successful there is the opportunity to: (1) improve patient care by detecting
common clinical conditions at an early stage, (2) avoid the need for advanced diagnostics in some cases, (3)
allow medical personnel to test various VAD settings to optimize output without the need for advanced
diagnostics, and (4) allow for remote diagnostics of VAD problems. In addition to Dr. Jason Kolodziej, Dr. Steven
Day, Dr. Linwei Wang, Dr. Karl Schwarz, Dr. Igor Gosev, and Dr. Michael Richards, the research team will consist
of three senior design teams (15 UG students), three to six undergraduate students, and a Ph.D. student.
