A centralised multi-objective model predictive control for biventricular assist devices / Vivian Koh Ci Ai

Vivian Koh , Ci Ai (2020) A centralised multi-objective model predictive control for biventricular assist devices / Vivian Koh Ci Ai. PhD thesis, Universiti Malaya.

	PDF (The Candidate's Agreement) Restricted to Repository staff only Download (216Kb)
Preview	PDF (Thesis PhD) Download (7Mb) | Preview

Abstract

Heart failure is defined as failure of heart to deliver adequate blood flow rate to support tissue perfusion. Heart failure can be treated by implantation of a left ventricular assist device (LVAD) for left heart failure patients, or a biventricular assist device (BiVAD) for bi-heart failure patients. Since left heart failure predominates right heart failure, all commercial ventricular assist devices are LVADs. Therefore, two LVADs are frequently used as BiVAD for bi-heart failure patients. Clinically, the constant speed (CS) control of BiVAD fails to adapt pump flow rate according to physiological changes, thus putting patients at risk of ventricular suction and pulmonary congestion. Speed regulation of a BiVAD may be complicated by process interactions in a cardiovascular-biventricular assist device (CVS-BiVAD) environment. Therefore, in this thesis, a centralised model predictive control (MPC) that could handle process interactions in a multivariable control problem was proposed. Three objectives were proposed in this thesis. Firstly, a simple state-space model of the CVS-BiVAD system was required prior to the development of an MPC algorithm. For this purpose, a complex CVS-BiVAD model was simplified by reducing the number of state variables. New model parameters were optimised using a least squares function and manual tuning approach. The simplified state-space model consists of state and time-varying factors. Therefore, the second objective was to modify a conventional centralised MPC algorithm to cater for the state and time-varying factors of the CVS-BiVAD system. Multiple control objectives were included in the MPC algorithm to: a) adapt pump flow rate according to the Frank-Starling (FS) mechanism, b)avoid ventricular suction, and c) avoid vascular congestion. This modified MPC iscalled the centralised multi-objective model predictive control (CMO-MPC). The CMO- MPC was benchmarked against two non-centralised control schemes: CS control and FS-like-proportional-integral (PI-FS) control under two patient scenarios: exercise and postural change, in silico, as the first stage evaluation. In exercise, CMO-MPC and PI-FS control increased cardiac output from 5.1 L/min to 7.1 L/min and 6.9 L/min, respectively. CMO-MPC avoided suction and congestion in both patient scenarios as compared to CS control and PI-FS control, based on the assumptions made on risks of suction (mean atrial pressure below 3 mmHg) and congestion events (mean atrial pressure above 18 mmHg). The assumptions only served as a proposed idea and can be changed by clinical experts. The third objective was to evaluate this CMO-MPC in-vitro using a mock circulation loop. In the in vitro study, CMO-MPC avoided pulmonary congestion in the exercise test while PI-FS control and CS control failed to. In the transient region of postural change test, CMO-MPC (2.0 mmHg) had a higher minimum right ventricular end diastolic pressure than PI-FS control (1.2 mmHg), suggesting that CMO-MPC had lower risks of right ventricular suction as compared with PI-FS control. It is therefore proposed that the CMO-MPC should be a safe physiological controller for BiVAD in the future when reliable pressure and flow sensors become clinically available. In vivo validation is also required to increase the confidence of use of CMO-MPC in the future.

Actions (For repository staff only : Login required)