Vorticity is not a metric a physician can currently order. There is no guide wire that measures it. It can be glimpsed, imperfectly, on a 4D flow MRI scan — but no patient is going to lie still in a scanner for the days or weeks needed to generate clinically actionable data. It is, for now, a computational biomarker: one that only exists inside the simulation.
“People don’t even have the same number of coronary arteries,” Randles notes. “If you run the exact same heart rate and the same flow rate through two different geometries, one can be completely colored with a high-risk state, and the other is completely fine. The geometry changes it so, so much.”
The result, at least in computational validation, is the ability to run what amounts to a continuous simulation of a patient’s cardiovascular state — tracking how blood flow metrics shift across weeks of ordinary life, including exercise, sleep, and recovery.
Her lab has already demonstrated proof-of-concept: using CT scans of patients’ pulmonary arteries and their computational flow simulations, they matched the pressure readings that CardioMEMS would have generated non-invasively. “We know that we can get the right answer,” she says.
The Carotid Trial: Predicting Recurrence Over Time
A second ongoing clinical study is tracking patients who have had revascularization procedures on their carotid arteries — procedures to remove or bypass dangerous arterial narrowing — and following them for up to a year using wearable sensors and periodic imaging.
The research question is longitudinal: can cumulative hemodynamic risk exposure, calculated from the flow simulation, predict how a patient’s artery will change at three-month and six-month follow-up visits? Can it identify disease recurrence earlier than current clinical markers?
“If we know your cumulative risk exposure over three months, is that going to help us predict any change in diameter faster and identify any recurrence of the disease easier?” Randles asks.
Results are not yet available, but the architecture of the question represents a meaningful departure from how vascular disease has been monitored: not a single-point measurement at a clinic visit, but an integrated risk score that accumulates over time and changes circumstantially.
Circulating Tumor Cells: An Unexpected Extension
The same simulation framework that models blood flow around an arterial lesion can, it turns out, model something else moving through those vessels: cancer cells.
Circulating tumor cells travel through arteries and veins until they lodge somewhere and metastasize. But what determines where they go? Why some cells are more likely to metastasize than others?
Randles’ platform offers a way to study the question with experimental precision. By introducing virtual cancer cells into the simulated blood flow — varying their shape, size, stiffness, and adhesive properties at the molecular level — the team can observe which characteristics cause a cell to spend more time near the vessel wall, which is a precondition for implanting.
“It really gives us a nice platform where we can keep everything absolutely the same and then just change, like, the radius of the cell very slightly and see if that affects where it’s going to go in the geometry or how much time it’s spending there,” Randles says.
The work focuses on solid tumors whose cells travel via the bloodstream, and the group is collaborating with microfluidics groups to characterize real cancer cell populations and validate the computational results against physical experiments.
The translational promise is therapeutic: if the simulation can identify what physical properties drive metastatic spread, it could help target drug development toward disrupting those specific mechanisms.
In all cases, the goal is not just to replicate what the implantable sensor already does, but to find better predictors — ones that could extend the warning window further or identify patients at risk earlier in their disease course. Having continuous, non-invasive access to pulmonary artery pressure data throughout the day, during exercise and recovery, opens a much larger discovery space.
“We’ve never had access to what your 3D blood flow looks like throughout the entire day,” Randles says. “We don’t know what the biomarkers are. Is it really going to be pulmonary artery pressure or is it the vorticity that changes three weeks ahead of time? Or the recovery after exercise and the change in your baseline from how vorticity may change after you exercise? Is that more predictive?”
Right now, the possibilities seem endless and the data problem daunting.
Running a 3D fluid simulation over weeks of continuous wearable data generates an enormous amount of information. Four and a half million heartbeats — roughly six weeks of continuous cardiac activity — have already been processed. The computational methods to scale simulations across those time horizons have been validated. What remains is the clinical question: in all that data, what patterns are actually predictive of disease events?
Randles suspects the answer will be personal in ways that population-level statistics cannot capture. A 5% change in post-exercise vorticity recovery might mean nothing for one patient’s geometry and signal serious trouble for another. Identifying meaningful phenotypic clusters — groups of patients whose anatomical and hemodynamic profiles are similar enough that shared thresholds make sense — may be the intermediate step between individual simulation and scalable clinical tool.
“Maybe there are 10 different clusters of patients where you can just figure out what is significant about your geometry and have some kind of phenotype,” she says. “But until we get all the data, it’s hard to tell.”
The trials will provide that data. The harder work, it seems, will be in making sense of it.