Application of Multiobjective Optimization to Determining an Optimal Left Ventricular Assist Device (LVAD) Pump speed
A Left Ventricular Assist Device (LVAD) is a mechanical pump used to assist the weakened left ventricle to pump blood to the entire body. One method of controlling pump speed is using a closed-loop controller that changes the pump speed based on the patients level of activity and demand for cardiac output. An important aspect of the development of a closed-loop controller is the selection of the desired pump speed. Pump speed must be chosen such that the patient receives adequate cardiac output for his/her level of activity. The pump must also operate in a safe physiological operating region, placing constraints on other hemodynamic parameters.
This work presents the pump speed selection problem as a multiobjective optimization problem, considering constraints on cardiac output, left atrial pressure, and arterial pressure. A penalty function is assigned to each hemodynamic variable and a mathematical model of the LVAD and cardiovascular system is used to map the penalty functions as functions of the hemodynamic parameters to penalty functions as functions of pump speed. The penalties for the different variables are combined by forming a weighted sum, and the best set of pump speeds is determined by minimizing the combined penalty functions using different sets of weights. The resulting set of best pump speeds forms the noninferior set (Zadeh, IEEE Trans. On Auto. Control, 1967). It was discovered that the noninferior set contains discontinuities, so the concept of a modified noninferior set known as the Clinicians noninferior set is introduced.
A decision support system (DSS) is presented that allows clinicians to determine a single pump speed from the noninferior set by investigating the effects of different speeds on the hemodynamic variables. The DSS is also a tool that can be utilized to help clinicians develop a better understanding of how to assign weights to the different hemodynamic variables.
Advisor:J. Robert Boston; Zhi-Hong Mao; Ching-Chung Li
School:University of Pittsburgh
School Location:USA - Pennsylvania
Source Type:Master's Thesis
Date of Publication:09/25/2007