Identification and Classification of Physiologically Significant Pumping States in an Implantable Rotary Blood Pump

Abstract

Abstract:In a clinical setting it is necessary to control the speed of rotary blood pumps used as left ventricular assist devices to prevent possible severe complications associated with over-or underpumping. The hypothesis is that by using only the noninvasive measure of instantaneous pump impeller speed to assess flow dynamics, it is possible to detect physiologically significant pumping states (without the need for additional implantable sensors). By varying pump speed in an animal model, five such states were identified: regurgitant pump flow, ventricular ejection (VE), nonopening of the aortic valve over the cardiac cycle (ANO), and partial collapse (intermittent and continuous) of the ventricle wall (PVC-I and PVC-C). These states are described in detail and a strategy for their noninvasive detection has been developed and validated using (n = 6) ex vivo porcine experiments. Employing a classification and regression tree, the strategy was able to detect pumping states with a high degree of sensitivity and specificity: state VE-99.2/100.0% (sensitivity/specificity); state ANO-100.0/100.0%; state PVC-I-95.7/91.2%; state PVC-C-69.7/98.7%. With a simplified binary scheme differentiating suction (PVC-I, PVC-C) and nonsuction (VE, ANO) states, both such states were detected with 100% sensitivity. Current commercially used implantable rotary blood pumps (iRBPs) make little or no attempt to automatically control pump speed to optimize ventricular assistance. In order to achieve such a control strategy, a major design goal for iRBPs is the ability to reliably and accurately detect pumping states that cause such deleterious effects as ventricular collapse due to overpumping, or pump backflow (regurgitation) as a result of underpumping (1). Naturally, the ideal control set point is where left ventricular (LV) ejection is occurring and there is a net positive flow through both the aortic valve (AV) and the pump.A state that would be of long-term concern to a patient would occur at higher relative pump speeds when there is insufficient blood in the ventricle to sustain normal LV ejection and the AV remains closed throughout the entire cardiac cycle. In this instance there is no flow through the AV and the possibility of blood stasis distal to the AV, which could lead to significant patient complications due to clotting. However, it has been recognized (2,3) that the aim of ensuring AV opening is often infeasible in patients with LV failure. The lack of native heart contractility in such patients means that in order to attain a level of pump flow which adequately perfuses the body, a pump speed which produces the state of full AV closure is often required. Even higher speeds would result in partial or total ventricular collapse as the volume of blood drawn from the ventricle chamber is increased to a level at which pulmonary supply cannot meet pump demand. Deleterious outcomes could also occur at lower relative pump speeds where there is regurgitant flow through the pump.Experimentation in the transition of pumping stat...

Left Ventricular Assist Devices: Engineering Design Considerations

“…These approaches use powerful pattern recognition algorithms to classify the signal into being in or out of suction. These classifiers vary from simple threshold comparisons (Oshikawa et al, 2000) to more complex techniques such as Classification and Regression Tree (CART) (Karantonis et al, 2006), Discriminant Analysis (DA) (Ferreira et al, 2006), and Neural Networks (NN) (Karantonis et al, 2008). In this chapter, we describe a new suction detection and classification algorithm based on the Lagrangian Support Vector Machine (LSVM) method in pattern recognition .…”

A Decision Support System for the Treatment of Patients with Ventricular Assist Device Support

“…Thus, some suction indices are based on time-domain features [54 -57], frequency-domain features [55 -57], and time-frequency-domain features [55 -57]. Among them, there exist methods which extract features from the pump flow signal being one of very few signals that can be easily measured and use powerful pattern recognition algorithms to classify the signal into different pump states [57][58][59][60][61]. However, in the majority of the above described approaches, a large number of features are employed.…”

A Dynamical Model for Pulsatile Flow Estimation in a Left Ventricular Assist Device

“…The experimental data were sampled at 200 Hz. For further information about data acquisition, see Karantonis et al (2006).…”

“…In each pig, the inflow pump cannula was inserted into the apex of the left ventricle while the outflow cannula was anastomosed to the ascending aorta (Karantonis et al, 2006). Indwelling catheters (DwellCath, Tuta Labs, Lane Cove, NSW, Australia) and pressure transducers (Datex-Ohmeda, Homebush, NSW, Australia) were instrumented to the pig's native heart to measure left ventricle, left atrial, aortic, and pump inlet pressures.…”