Collaborative research: FDT-biotech: coupled patient-circuit digital twin of cardiopulmonary bypass for uncertainty-aware in-silico evaluation of automated perfusion

The National Science Foundation Directorate for Mathematical and Physical Sciences awarded Vanderbilt University Medical Center $491,748 on October 1, 2026, under the Mathematical and Physical Sciences program (CFDA 47.049) to develop an uncertainty-aware digital twin that couples patient and cardiopulmonary bypass machine dynamics for in-silico evaluation of automated perfusion control.

The project will create a computer-based replica of the patient and heart-lung machine that updates as new clinical information becomes available, estimates hidden indicators of oxygen delivery, and quantifies uncertainty in those estimates. The model combines established physiological and engineering knowledge with constrained machine learning from clinical data to ensure predictions remain physically and medically plausible. The work will support safe computer-based testing of automated approaches for managing blood flow and oxygen delivery before clinical use, including rare and high-risk scenarios difficult to study in trials. The project will use data from adult and pediatric heart surgeries across multiple hospitals, release openly available software and privacy-preserving collections of simulated patients, and develop evaluation practices to inform future regulatory review. The resulting tools could improve safety and efficiency in heart surgery and may extend to other systems supporting the heart or lungs, as well as automated drug delivery. The project will also train students across computing, engineering, mathematics, and medicine and create new educational materials and an interdisciplinary workshop.

Work is performed in Nashville, Tennessee, with a period of performance from October 1, 2026, through September 30, 2029.

Abstract

Each year, heart-lung machines support hundreds of thousands of patients in the united states during complex heart operations by temporarily taking over the work of circulating and oxygenating the blood. Clinicians must continuously adjust these machines using measurements that arrive at different times and may not clearly reveal whether the patient?s organs are receiving enough oxygen. These limitations can contribute to complications such as acute kidney injury. This project will create a computer-based replica of the patient and heart-lung machine that updates as new information becomes available, estimates hidden indicators of oxygen delivery, and reports how uncertain those estimates are. The model will combine established knowledge of the body and machine with carefully constrained learning from clinical data, so that its predictions remain physically and medically plausible. IT will support safe computer-based testing of automated approaches for managing blood flow and oxygen delivery before any use in patients, including rare and high-risk situations that are difficult to study in clinical trials. The project will use data from adult and pediatric heart surgeries across multiple hospitals, release openly available software and privacy-preserving collections of simulated patients and develop evaluation practices that can inform future regulatory review. The resulting tools could improve the safety and efficiency of heart surgery and may benefit other systems that temporarily support the heart or lungs, as well as automated drug delivery. The project will also train students across computing, engineering, mathematics, and medicine and will create new educational materials and an interdisciplinary workshop. The project develops an uncertainty-aware digital twin for coupled patient-device dynamics during cardiopulmonary bypass. Its mechanistic backbone is the pulse physiology engine, augmented by constraint-projected residual stochastic differential equations that learn patient-specific model mismatch while preserving conservation laws, positivity, phase-specific operating constraints, and identifiability of mechanistic parameters. A constraint-projected nonlinear filter will jointly estimate latent physiologic states, slowly varying parameters, sensor biases, and delays from noisy, asynchronous, and out-of-sequence observations, with posterior uncertainty restricted to the same admissible manifold as the dynamics. The estimated belief state will drive a distributionally robust, risk-sensitive model predictive controller that minimizes oxygen-delivery deviations and cumulative oxygen debt subject to safety and chance constraints. Its wasserstein ambiguity radius will be calibrated by a new phase-aligned optimal-transport validation framework that compares distributions of real and simulated trajectories while controlling for nuisance timing and sampling variation. Full and sliced wasserstein variants will provide cohort-scale metrics, interpretable discrepancy localization, and stability guarantees. The integrated framework will be evaluated on harmonized adult, pediatric, and multi-institutional perfusion data through uncertainty-aware digital cohorts and a preregistered controller benchmark against rule-based and learned baselines. Deliverables include open-source modeling, filtering, control, and validation software; privacy-preserving cohort-generation tools; and verification, validation, and uncertainty-quantification artifacts aligned with recognized standards for precompetitive engagement with federal regulators. This project is jointly supported by the national institutes of health through the office of data science strategy and the national institute of biomedical imaging and bioengineering. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the foundation's intellectual merit and broader impacts review criteria.- subawards are not planned for this award.