Thumbnail Image


Publication or External Link






Automated infusion of medications will be increasingly deployed in patient care as a means to deliver high-quality and continuous monitoring and therapy, and also to alleviate the excessive workload imposed on the clinicians. Therefore, a well-designed automated medication infusion system is an attractive alternative to today’s manual treatment requiring caregiver’s interventions. However, it also presents numerous challenges: 1) Significant inter- and intra-patient variability; 2) Complexity of medication infusion model; 3) Complexity of interaction of multiple medications; 4) Difficulty in coordination of medical targets.

So the following approaches are proposed to address the various challenges:

First, to deal with the large degree of individual patient variability, an adaptive controller was designed. This is because robust controllers which have fixed parameters might be difficult to offer decent behavior for all patients.

Secondly, since classical adaptive controllers can only be applied to linearly parametrized models while even the infusion model of single drug is highly nonlinear and complex, a single-input single-output (SISO) semi-adaptive control approach which only adapt can adapt model parameters having a large impact on the model’s fidelity was introduced.

Thirdly, the complicated interaction of multiple medications makes the adaptive controller for two medications even more difficult to design. So a model for two interacting dose responses was constructed and linearized at one operation point. Then the SISO semi-adaptive controller was extended to a two-input two-output case. However, this controller is only designed at one operating point. Therefore, based on two models associated with two distinct operating regimes, a two-model switching control technique was developed and combined with the semi-adaptive controller.

Fourthly, we presented a coordinate mechanism to deal with the medical targets setting problem. In real clinical scenarios, the reference targets are empirically specified by caregivers, which are not always achievable in all patients. Therefore, our proposed coordinate mechanism can recursively adjusts the reference targets based on the estimated dose-response relationship of a patient.

Lastly, we conducted some SISO control experiments on animals. Based on the experiments, we made some further improvements to the proposed controller.