Albers, D. J. | bioRxiv Science

bioRxiv2025

Physics-Informed Digital Twin Can Predict Cerebral Blood Flow and Cerebral Vascular Regulation Mechanisms in Neurocritical Care Patients

1Cerebral blood flow is vital for brain function and is acutely controlled through a set of physiological mechanisms known as cerebral vascular regulation (CVR)[1-3]. It remains challenging to directly measure the dynamics and function of individual CVR mechanisms[4], limiting our ability to understand and optimize brain perfusion, particularly for neurologically injured patients. Digital twins offer an ideal tool for overcoming this gap because they enable estimation, tracking, and forecasting of unmeasured physiological states[5, 6]. Here, we introduce CereBRLSIM (Cerebral Blood Regulation Latent State Inference and Modeling), a digital twin that integrates physiological knowledge and patient data to infer CVR function and predict cerebral dynamics. Using both in vivo experiments and simulated data, CereBRLSIM predicted cerebral hemodynamics with high accuracy and estimated the dynamics of myogenic, endothelial, and metabolic mechanisms underlying CVR. When personalized to neurocritical care patient data, CereBRLSIM differentiated cerebral hemodynamic phenotypes, predicted patient outcomes, and forecasted blood flow with significantly higher accuracy than machine learning models. This work provides a novel, interpretable, and clinically compatible approach for quantifying CVR function and forecasting cerebral blood flow, enabling new opportunities in precision diagnostics and foundational understanding of cerebral hemodynamics.

bioengineering↗

bioRxiv2020

A damaged-informed lung model for ventilator waveforms

Motivated by desire to understand pulmonary physiology and pathophysiology, scientists have developed models of pulmonary physiology. However, pathophysiology and interactions between human lungs and ventilators, e.g., ventilator-induced lung injury (VILI), present problems for modeling efforts. Real-world injury is too complex for simple models to capture, and while complex models tend not to be estimable with clinical data, limiting both the clinical utility with existing approaches. To address this gap, we present a damaged-informed lung ventilator (DILV) model to model and quantify patient-ventilator interactions and lung health. This approach relies on systematically mathematizing the pathophysiologic knowledge clinicians use to interpret lung condition from ventilator waveform data. This is achieved by defining clinically relevant features in the ventilator waveform data that contain hypothesis-driven information about pulmonary physiology, patient-ventilator interaction, and ventilator settings. To capture these features, we develop a modelling framework where the model has enough flexibility to reproduce commonly observed variability in waveform data. We infer the model parameters with clinical (human) and laboratory (mouse) data. The DILV model can reproduce essential dynamics of differently damaged lungs for tightly controlled measurements in mice and uncontrolled human intensive care unit data in the absence and presence of respiratory effort. Estimated parameters correlate with known measures of lung physiology, including lung compliance. This method has the potential to translate laboratory physiology experiments to clinical applications, including pathways for high fidelity estimates of lung state and sources of VILI with an end goal of reducing the impact of VILI and acute respiratory distress syndrome.

physiology↗