Model Predictive Control for Renal Anemia Treatment through Physics-informed Neural Network

Introduction:

Predicting hemoglobin is useful, but treatment also requires deciding how much erythropoietin to give and when to give it. This paper builds that decision-making step using zone model predictive control. Instead of forcing hemoglobin toward one exact value, the controller aims to keep it within a clinically acceptable range. The prediction model is a physics-informed neural network for control, which can take time, current patient state, and future dosing decisions as inputs. This makes it suitable for rolling treatment optimization. By combining physiology-guided learning with control design, the method provides a simulation-based decision-support tool for individualized renal anaemia treatment.