# Physics‑based simulations help diagnose and treat disease

An episode of [Physics World Weekly Podcast](https://invisiblenetwork.ai/podcast/1400686733), 5 Feb 2026, 28 min.

Audio: https://media.blubrry.com/physicsworldweeklypodcast/content.blubrry.com/physicsworldweeklypodcast/PWWeekly5Feb2026AmandaRandles.mp3

Computer scientist Amanda Randles is our podcast guest

## Transcript

[0:08] Hello, and welcome to the Physics World weekly
[0:11] podcast.
[0:12] I'm Hamish Johnston.
[0:14] This episode
[0:15] features
[0:16] Amanda Randalls,
[0:17] who is a computer scientist and biomedical
[0:21] engineer
[0:22] at Duke University
[0:23] in The
[0:25] US. Duke is also where she began her
[0:28] scientific career
[0:29] by studying for an undergraduate degree in physics.
[0:34] For the past few years, she's been using
[0:37] physics based
[0:38] computationally
[0:39] intense
[0:40] simulations
[0:41] of the circulatory
[0:42] system
[0:43] to identify signs of heart disease
[0:46] before symptoms start.
[0:48] These are signs that show up not only
[0:51] in clinical tests, but also in data from
[0:55] wearable devices,
[0:56] such as smartwatches
[0:58] that record information
[1:00] while patients are going about their daily lives.
[1:04] More recently, Randalls has also become interested
[1:08] in understanding
[1:09] how cancer cells
[1:11] move through the bloodstream
[1:13] and initiate
[1:14] metastasis,
[1:16] which is the process where cancer that started
[1:19] in one location within the body
[1:21] spreads to other organs.
[1:24] In 2024,
[1:26] the Association for Computing Machinery
[1:29] awarded Randalls its ACM prize in computing
[1:33] for her groundbreaking
[1:35] contributions to algorithms
[1:37] and high performance
[1:39] computing methods
[1:40] aimed at diagnosing
[1:42] and treating
[1:43] human diseases.
[1:45] As well as scientific
[1:47] prestige
[1:48] and $250,000
[1:51] in cash,
[1:52] This prize comes with an invitation
[1:55] to the annual
[1:56] Heidelberg
[1:57] Laureate Forum,
[1:58] or HLF,
[2:00] which brings prize winning researchers
[2:03] and early career researchers
[2:05] in computer science and mathematics
[2:08] to Heidelberg,
[2:09] Germany
[2:10] for a week of talks and networking.
[2:14] My colleague Margaret Harris caught up with Randalls
[2:17] at last year's h l f.
[2:20] But interestingly,
[2:21] this wasn't Randalls first trip to the forum.
[2:24] She'd previously attended
[2:26] when she was an early career researcher
[2:29] herself.
[2:30] You'll hear her talk about this experience later
[2:33] in the podcast,
[2:34] But
[2:35] first, let's hear about her early interest in
[2:38] physics
[2:38] and how it fed into a concept
[2:41] that's known as a digital twin.
[2:52] My first question is, how did you get
[2:54] interested in physics in the first place?
[2:58] Yeah. I I think basic things of, like,
[2:59] you know, I was always a kid, you
[3:01] know, who was always asking lots of questions.
[3:03] And my parents ended up buying me, like,
[3:05] that whole set of the, like, ask me
[3:06] why books to, like, you know, try to
[3:09] address that.
[3:10] In high school, I went to a math
[3:11] and science center for half the day every
[3:13] day for, the high school programming. And that
[3:15] was where I, you know, first took some
[3:17] physics classes and really kind of fell in
[3:18] love with it and and really liked it.
[3:20] I had like, the the physics teacher there
[3:22] was fantastic and and just, I don't know,
[3:24] made everything really exciting.
[3:26] So I knew when I went to Duke,
[3:27] I really wanted to study some kinda like
[3:29] biophysics or or something in that space.
[3:32] And then, yeah, I just I I really
[3:33] liked it when I when I was there.
[3:34] I I I like the, you know, the
[3:36] approach of really, you know, how do you
[3:37] understand something from first principles and and really
[3:40] get to the answer. And And you're continuing
[3:42] to apply that sort of physics based approach
[3:44] to your work now, which is very much
[3:46] on the board between
[3:47] computer science, biomedical engineering,
[3:50] and particularly this concept of the digital twin.
[3:52] I wonder if you could just tell our
[3:53] listeners what a digital twin is.
[3:57] Yeah. So so, basically, like, we're we're reviewing
[3:59] a digital twin is where it's a virtual
[4:01] replica of a physical entity that's gonna be
[4:04] continuously updated with information driving, like, from that
[4:07] physical entity. So you have, you know, potentially
[4:09] sensors are gonna be giving continuous feedback to
[4:12] that entity and then allowing you to draw
[4:14] something that may, you know, then influence that
[4:16] entity in the future.
[4:17] I think this is an idea that actually
[4:19] has a quite long history in engineering. So,
[4:22] yeah. This this idea is it's very popular
[4:24] right now, but it's not it's it's not
[4:26] anything new. They've been doing it for years.
[4:28] It's been really popular in civil engineering for
[4:30] a long time. You see it a lot
[4:32] in mechanical engineering. So there there's constantly
[4:34] you know, we see strong examples of using,
[4:36] you know, digital twins of bridges where you
[4:38] have sensors all over the bridges to see,
[4:40] like, what the load is like and getting
[4:41] that information about, like, that specific bridge. They
[4:44] had the same thing. You know, many companies
[4:45] have been doing this for their jet engines
[4:47] for years.
[4:48] We're even starting to see it for, like,
[4:50] supercomputing centers. We have a digital twin at
[4:52] the supercomputing center to, you know, watch, you
[4:54] know, the power consumption, cooling, and that that
[4:56] side of things. So it's pretty pervasive. We're
[4:58] just we're we're trying to kind of make
[5:00] use of these advances and use it for
[5:02] the health side and start trying to make
[5:03] this digital twin of the human. It it's
[5:05] been around for a while. Could you give
[5:07] me an example of what's a digital twin
[5:09] might how it might work in practice in
[5:10] your field?
[5:12] So we're we're really starting with just focusing
[5:14] on a digital twin of vascular of a
[5:16] like a vascular digital twin. So we're looking
[5:18] at your circulatory system, and we're building a
[5:20] replica. So we need medical imaging. We get
[5:23] a three d model of your, you know,
[5:25] coronary arteries, for example.
[5:27] And
[5:28] in that case, we then can extract you
[5:31] know, we get the three d geometry representing
[5:33] that specific patient's coronary arteries, and then we
[5:36] get information from, like, wearable devices. So we
[5:38] can use that to understand, you know, what
[5:40] is your heart rate, kind of try to
[5:42] figure out what is what are the flow
[5:43] properties, and try to assess, you know, what
[5:45] would be the velocity going into those coronary
[5:47] arteries and drive the flow simulations from the
[5:49] data that we're able to get from the
[5:51] wearables. And then you could imagine as someone's,
[5:53] you know, going through their daily life,
[5:55] typically, what a doctor would look at is,
[5:57] you know, information they can get in the
[5:58] doctor's office, what they can measure in real
[6:00] time over just a couple heartbeats.
[6:03] And what we're trying to do is understand
[6:05] how is someone's three d blood flow changing
[6:08] as they go about, you know, a week
[6:09] or a month to see, you know, not
[6:11] just, like, did you enter a vulnerable state,
[6:13] but, you know, how long did you spend
[6:15] in these vulnerable states, and is that helping
[6:16] us predict or identify
[6:18] identify any kind of disease progressions? You could
[6:20] imagine, you know, someone wearing this for heart
[6:23] failure and before they ever have symptoms of,
[6:25] like, you know, shortness of breath or kind
[6:28] of chest pain, there may be a change
[6:30] in their hemodynamics
[6:31] that the digital twin can identify.
[6:33] And then the doctor might be able to,
[6:34] like, you know, see that that change has
[6:35] happened and either, you know, change their medication,
[6:38] ask them to come in, do something that's
[6:40] a bit more proactive. So kind of use
[6:41] it as a way of monitoring the patient
[6:43] while they're at home. And I guess one
[6:45] of the things that connects
[6:46] the examples you gave in civil engineering and
[6:48] what you're now trying to do with humans
[6:51] and health care
[6:52] is that you can't do
[6:54] infillimen
[6:55] experiments
[6:56] on other system, and you can't do things
[6:58] to the bridge that's gonna destroy it, just
[7:00] like you can't do anything to a human
[7:01] that's going to in the severely hormone, unless
[7:04] there's a commensurate benefit, like you're doing in
[7:06] face of test, for example. Right. So how
[7:08] does it work to sort of link
[7:11] what what's the time scales on which you're
[7:13] linking the information you're gathering from wearables, you're
[7:15] gathering from scans, and how are you up
[7:16] using it to update the digital twin so
[7:18] you get information about the the actual work?
[7:21] Yeah. I know. That's a great question. So
[7:22] the way the way we're doing it is
[7:24] that we really see it as, you know,
[7:25] a blending of data from things like the
[7:26] wearable sensors with the physics based modeling. What
[7:29] we're trying to set up right now is
[7:31] we're we've developed this method. We call it
[7:32] longitudinal hemodynamic mapping framework, and it allows you
[7:35] to really kind of take a parallel on
[7:37] time approach to simulating the hemodynamics.
[7:40] And it allowing like, basically, we're we're we're
[7:42] getting at, you know, a finite number of
[7:43] hemodynamic units that we can identify that if
[7:46] you can piece those back together and and,
[7:48] you know, figure out it's almost like a
[7:49] lookup table of, you know, you figure out
[7:51] what these hemodynamic units are. You use your
[7:53] wearable data and you figure out which hemodynamic
[7:54] unit you should be pulling from. What we
[7:57] need to do then is simulate all of
[7:58] those hemodynamic units when we first get the
[8:00] medical imaging. So we're trying to minimize the
[8:02] amount of time it takes, you know, the
[8:03] number of hemodynamic units and how efficiently we
[8:06] can calculate each of those. But by doing
[8:08] that, you know, it might take
[8:10] somewhere in the order of, like, fifteen minutes
[8:11] to half an hour. If it's a really
[8:13] complicated geometry, maybe a little bit longer to
[8:15] segment the geometry
[8:16] and get the data, like, you know, get
[8:17] that geometry from the medical imaging. Once you
[8:19] have that, then it can take, you know,
[8:21] a few hours to maybe a day or
[8:23] two,
[8:24] to run all of those hemodynamic units. But
[8:27] then the goal is then in practice is
[8:28] once you've run the simulations,
[8:30] it works like a lookup table. You can
[8:32] then use that. So then, you know, it
[8:34] could almost run on the on your devices
[8:36] or on edge computing,
[8:37] setups
[8:38] to have, you know, draw driven from your
[8:40] wearable device to identify what are those hemodynamic
[8:42] units and then have algorithms on that that
[8:44] can can kind of track and see what
[8:46] the the markers are. And that's, you know,
[8:48] how we would see it for diagnostics or
[8:50] for tracking.
[8:51] But then there's, you know, the other side
[8:53] of things where, you know, that's assuming we
[8:55] know what the marker is that we're looking
[8:56] for. There's a whole another piece where it's
[8:58] trying to just, you know, do a discovery
[9:00] phase where we wanna create, save all of
[9:02] the three d blood flow over six months
[9:04] of time, do a large scale, you know,
[9:07] analysis to identify what is that marker that
[9:09] is changing before you have shortness of breath,
[9:12] and can we identify what that is? So,
[9:14] you know, that takes much longer and actually
[9:15] needs, you know, full simulations, all of the
[9:17] data retained.
[9:19] But that allows you to then, you know,
[9:20] create potential AI surrogates to go on the
[9:22] edge and pieces like that. So there's there's
[9:23] kinda two pieces of when you would wanna
[9:25] deploy it and how you'd really use it
[9:27] to find new markers.
[9:28] And I can see so you're talking about
[9:30] simulating these little chunks of a bloodstream effectively.
[9:33] It's not not quite sort of individual red
[9:35] blood cells, but are are are chunks of
[9:37] the bloodstream.
[9:38] How much computing power does that currently take
[9:41] and what challenges does that
[9:44] present? Like, I mean, a lot of our
[9:45] goal is to reduce the computing power. So
[9:48] a lot of this is an interesting,
[9:50] we have to do, like, the brute force
[9:51] simulation
[9:52] to make sure we're, like, you know, we're
[9:54] keeping everything in that we need to. So
[9:56] we're it requires a lot of times, like,
[9:58] the world's largest supercomputers,
[10:00] but that's only to run, like, the validation
[10:02] ground truth simulation that we're then using to
[10:04] test our algorithms against. So we're we're working
[10:06] you know, we have a lot of different
[10:07] models to go, you know, longer in time
[10:09] or deeper,
[10:10] spatially
[10:12] that, you know, we run the huge simulation
[10:13] on the large supercomputer, and then we'll run,
[10:15] you know, on several nodes in the cloud
[10:17] and compare against that simulation.
[10:19] So to to really, you know, put it
[10:21] into use, you would only need the few
[10:22] nodes on the cloud. But it's, you know,
[10:24] critically important to do the testing. And should
[10:26] note, like, we're pairing that testing with, you
[10:28] know, testing with Doppler ultrasound and testing with,
[10:30] you know, implantable devices. So also trying to
[10:33] make sure it's actually matching the patients as
[10:35] well in the experimental side, but I think
[10:38] having that verification with, like, the ground truth
[10:40] has been important. So, yeah, we we we
[10:42] do use a lot of the, yeah, like,
[10:44] the worst biggest supercomputers to run some of
[10:46] these simulations. And our goal is to try
[10:48] to, you know, put absolutely everything into the
[10:49] model so we can figure out what we
[10:51] can take out and how to pair it
[10:52] down. And so you do that book out
[10:53] table every time. Yes.
[10:55] Yeah. What are some principles that's
[10:58] that you're using in your work you're looking
[11:00] for in when you're simulating,
[11:02] the circulatory system? Yeah. So I guess, fundamentally,
[11:05] I think the the key piece is, like,
[11:06] it's not just we're not just looking at,
[11:08] like, random data points. There's not, like, just
[11:10] a data driven,
[11:11] approach. Like, this really truly is a physics
[11:14] informed,
[11:15] like, we are running physics based modeling,
[11:17] where you are getting less, like, three d
[11:19] blood flow model.
[11:20] We are taking that three d model that
[11:22] you're getting out of the the medical imaging
[11:24] from segmentation.
[11:25] We use a discretized approach. So we put,
[11:27] like, a regular like, just a regular Cartesian
[11:29] grid across that mesh.
[11:31] We figure out, like, what's an inlet node,
[11:32] what's an outlet node, and then you're just
[11:33] solving, you know, very basic fluid dynamics equations
[11:36] at each of those grid points.
[11:38] It's really, really, you know, very fine mesh,
[11:41] so you need, like, you know, so many
[11:42] grid points you end up needing large scale
[11:44] supercomputers.
[11:45] But at the end of the day, you're
[11:46] just solving you know, we use a lot
[11:47] of Boltzmann, but it's a way of solving
[11:48] Navier Stokes. So we're really just solving, you
[11:50] know, basic fluid equations at each of these
[11:52] at each of these grid points. And that
[11:54] allows us to recover,
[11:56] you know, the three d float profile.
[11:58] So we can get quantities like vorticity, velocity,
[12:00] we need pressure, and we can get that
[12:02] in a three d format. And then we're
[12:04] combining that with cellular models. So we have
[12:07] finite element models of, you know, red blood
[12:09] cells as they're moving through the geometry, how
[12:11] they're interacting with that fluid.
[12:13] But it is it's important to have, like,
[12:15] a physics understanding and, like, a physics based
[12:17] model of, like, how is that red blood
[12:19] cell deforming as it's moving through? Where is
[12:22] it going to adhere? How is it interacting?
[12:24] And it it it's all coming from the
[12:25] physics.
[12:26] I think that's been your your focus for
[12:28] for several years now. Yeah. And but you're
[12:31] now actually starting to shift as well while
[12:33] still working on that to
[12:35] understanding, like, the biology and the physics impact
[12:38] of of cancer and what that looks like
[12:40] in a digital twin sort of simulation.
[12:42] What's that look like for you? Yeah. So
[12:44] everything we've done on, like, the heart disease
[12:46] side, you really don't need to have individual
[12:49] cells in that case. Like, having bulk fluid,
[12:51] you know, is enough to recover, like, the
[12:53] values that we're interested in that are known
[12:55] biomarkers for heart disease. But by having this
[12:58] large scale,
[12:59] kind of framework, you know, we started adding
[13:01] in red blood cells probably I don't know.
[13:03] I think, like, 2,009, 2,010.
[13:05] And, we realized you could use this model
[13:08] for studying cancer and trying to understand
[13:11] we we don't look at tumors themselves. We
[13:12] tend to look at when a cancer cell
[13:14] has ripped off a primary tumor cell and
[13:16] is moving through the bloodstream.
[13:18] We wanna understand what it is about that
[13:20] cancer cell that is making it choose one
[13:22] path or the other path. Why is it
[13:24] spending so much time at certain locations on
[13:26] the wall? What's gonna make it more likely
[13:28] to adhere to that location on the wall?
[13:31] So from that end, it's, you know, it
[13:32] it's an incredibly computationally intense but very physics
[13:35] based question. So we, you know, we're trying
[13:37] to build
[13:38] accurate models of the cell membrane so we
[13:40] can, you know, compare to microfluidic devices and
[13:43] make sure, you know, we're getting those specific
[13:44] cancer cells accurately.
[13:46] We're trying to add we have models of,
[13:48] like, ligand receptor interactions. So we have you
[13:50] know, one one of the projects we have
[13:51] going right now is,
[13:53] having the large scale fluid model to understand
[13:55] the wall shear stress on, like, the vascular
[13:57] like, on the endothelium and on the walls,
[13:59] and then using that to drive these the
[14:01] receptor expression and understand how that's, you know,
[14:04] influencing the endothelium and causing different receptors. Like,
[14:07] trying to, like, predict what that pattern is
[14:09] going to be based on that wall shear
[14:10] stress interaction.
[14:12] And then you have the computational model of
[14:14] the cancer cell moving through to see where
[14:15] is it going to here and really try
[14:17] to, you know, find a way of parsing
[14:19] the dynamics of, you know, the fluid combination
[14:22] and how it's affecting the cells and then
[14:24] how the cells are interacting in that space
[14:26] and giving you this way. The computational tools
[14:29] are fantastic and that, you know, we can
[14:31] control absolutely everything. You can leave everything exactly
[14:34] the same and tweak just, you know, the
[14:36] velocity slightly or the bond strength slightly. And
[14:39] then you can really isolate, you know, how
[14:41] did that bond strength change
[14:43] the amount of time you spent at the
[14:44] wall or your residence time for that cancer
[14:46] cell. And it really lets us
[14:48] fine tune test these items that it's harder
[14:50] to do in an experiment when, you know,
[14:52] humidity might be changing. Or, like, you know,
[14:54] there are many other quantities that are changing
[14:56] that you can't control. So it offers, you
[14:58] know, a nice compliment to the experiments in
[14:59] that way. And are you exclusively interested in
[15:02] this
[15:03] metastasis process whereby a a cell or a
[15:06] cancer cell breaks off from the initial tumor
[15:08] then travels to somewhere else in the bloodstream
[15:11] to produce a cancer somewhere else? Is that
[15:13] is that the main how they're focusing it?
[15:14] Yeah. That that's really what we've looked at
[15:16] at this point. We're we're starting to branch
[15:18] a little bit into like, you know, we
[15:19] we model all the red blood cells. We're
[15:20] kind of branching a little bit into, you
[15:22] know, how to what changes with sickle cell
[15:23] anemia,
[15:24] like, you know, changes to the red blood
[15:26] cells themselves.
[15:27] And even just as you age, how do
[15:29] the mechanical properties of the red blood cells
[15:31] change, and how does that affect, you know,
[15:33] your response to diseases? So we're kind of
[15:35] branching into, like, the red blood cell side
[15:37] as well, but a a lot of the
[15:39] focus has been just on the can we
[15:40] understand what it is about the cancer cell
[15:42] that's driving like, the mechanics of the cancer
[15:44] cell that's driving cancer metastasis or the likelihood
[15:47] of disease progression.
[15:49] Thinking about challenges now as you try to
[15:52] move forward in research, what would you say
[15:54] the biggest
[15:55] physics related challenges
[15:57] are?
[15:58] Yeah. I think a lot of it has
[16:00] been, trying to make stable simulations that can
[16:03] cross spatial and temporal boundaries. So
[16:06] for the cancer project,
[16:07] we want to understand
[16:09] it's not just the deformation of the cancer
[16:11] cell or the nucleus which requires a high
[16:13] resolution.
[16:14] But then we really want these ligand receptor
[16:15] pairings, but we wanna understand how it how
[16:18] those, you know, adhesion is changing the behavior
[16:21] over a centimeter scale or, you know, it'd
[16:22] be great if we get a mill like,
[16:24] a a meter scale. But by doing that,
[16:26] you you know, we've been using, like, this
[16:28] approach we call, like, adaptive physics refinement. It's
[16:30] very similar in spirit to, like, adaptive mesh
[16:32] refinement.
[16:33] But as soon as you go in that
[16:34] direction, it's like, you know, how do you
[16:35] stably couple these different levels of different mesh
[16:38] refinements and how do you get you know,
[16:39] in our case, we have one physics model
[16:41] being used at the the fine grain model,
[16:43] the coarse grain model, a different physics unit.
[16:45] And how do you have that couple in
[16:47] a stable way that is, you know, accurate
[16:49] and continuing?
[16:50] We have, like, in the in the coarse
[16:52] grain model, it's the bulk fluid, which has
[16:54] a very viscous fluid. So from the even
[16:56] just the fluid side, it's coupling to, like,
[16:58] the plasma, which is more like a water.
[17:00] So you're you're introducing it just a lot
[17:02] of points of potential instability, and it's, it's
[17:04] requiring a lot more on the physics and
[17:06] the algorithm side versus just, you know, we
[17:08] know the equation. Let's try to make it
[17:09] faster. And there's there are current limitations in
[17:12] what the resolution gap between that fine grain
[17:15] and the coarse grain model can be. And
[17:17] the more we can push that boundary and
[17:19] allow that to widen would really help us.
[17:22] And in terms of computational challenges, what are
[17:24] the big issues there? I think at the
[17:26] point at this point, like, data is a
[17:27] huge issue of, you know, we I keep
[17:29] saying, like, we're really great at, you know,
[17:31] the input, the STL file that you're getting
[17:33] at the triangulated mesh is
[17:34] probably a couple of kilobytes.
[17:36] You're getting, you know, a few your input
[17:38] file is, like, 10 numbers in it. It's,
[17:40] you know, it's very small data. And then
[17:41] we are running on the world's biggest supercomputers
[17:44] and using all of the data, which can
[17:46] be on the order of a petabyte of
[17:47] data for one time step. And then you're
[17:49] trying to run for millions and millions of
[17:51] time steps. So, you know,
[17:54] we're generating
[17:55] petabytes and petabytes of data. Obviously, we can't
[17:58] store all of that data. So it's, you
[17:59] know, how do you analyze that data? How
[18:01] do you make use of it? What do
[18:02] you store?
[18:03] It's like data has really become
[18:05] a a huge part of it. Like, we're
[18:07] we're doing a lot of work on in
[18:08] situ machine learning and in situ visualization and
[18:10] what can we do while the data is
[18:11] still in memory on the super computer. But
[18:14] it just trying to figure out, you know,
[18:15] and that's, you know, just for one patient,
[18:17] if you wanna start looking across multiple patients,
[18:19] how do you bring all this together and
[18:20] actually analyze it in a way that that's
[18:22] tractable and useful?
[18:23] It's I think we we've had a lot
[18:25] of success scaling the code, which is
[18:31] is great, but that has then led to
[18:31] this problem of now we have tons of
[18:31] data, which is, you know, a good problem
[18:31] to have, but it's it's definitely an issue.
[18:33] Touch the part more businesses. I know. Yeah.
[18:35] It's
[18:37] So part about this, what is something that
[18:39] you can't do now
[18:41] that you see coming in the future and
[18:44] you're really excited about?
[18:46] Right now,
[18:48] for the for the right now, for the
[18:49] digital twins, especially even just on the heart
[18:51] side,
[18:52] we've done a lot of work
[18:54] validating that, like, at the we can match
[18:56] single time point measures of what you get
[18:57] in the clinic. And then that's, you know,
[18:59] a critical first step, really important.
[19:01] But what we want to see happen is
[19:03] that we are getting the right data as
[19:05] you go out into the world and you're
[19:06] wearing your wearable and we're getting the right
[19:08] data. And it it looks promising, but we
[19:10] have a lot of studies going on to,
[19:11] like, validate that data. You know, we've been
[19:13] doing a lot of comparisons against him for
[19:15] heart failure for the an implantable device that
[19:18] they it measures once a day when you're
[19:19] laying down, when you're sedentary.
[19:21] And we can match that very strongly, and
[19:23] it looks really promising. But now we need
[19:25] to do it while you're exercising, while you're
[19:27] moving around and see can we get heart
[19:29] rate recovery. Are we really capturing what's happening
[19:32] throughout the day and dynamically? And I think
[19:34] once we can do that,
[19:36] it really opens up that's where you're gonna
[19:37] find the new biomarkers, and that's where you're
[19:39] gonna be able to identify something that goes
[19:40] further than what we have now. And it's
[19:42] you know, we have tons of collaborations in
[19:45] place to get the data we need from
[19:46] wearables to be able to make that jump,
[19:49] But it's still an open question of, like,
[19:50] what's gonna be the right way of measuring?
[19:52] Like, you know, for instance, we need stroke
[19:53] volume. That is an incredibly hard measurement to
[19:56] get with wearables, and we have many projects
[19:59] trying to figure out how to do that
[20:00] accurately. But it's until you really have that
[20:03] estimation in there, it it's incredibly promising, but
[20:06] there's still work to be done. So stroke
[20:08] volume is how much blood gets pulse when
[20:10] this with a single heartbeat. Yes. Yep. So
[20:12] we can get heart rate easily, but we
[20:14] really need cardiac output. And for that, you
[20:15] need heart rate and stroke volume. And, Usually,
[20:18] like, a tiny robot inside of the body,
[20:20] kind of measuring things inside the the artery
[20:22] or something. Yes. Yep. Yeah.
[20:25] That would be really helpful.
[20:27] And talk a little bit about how the
[20:29] physics that you use in your work. Which
[20:30] aspect of physics would you say that you
[20:32] use most? What lesson from your physics education
[20:35] do you do to keep coming back to?
[20:38] I feel like it's more just like the
[20:40] approach of
[20:41] really, you know, trying to under like, trying
[20:43] to understand what's going on from the fundamental.
[20:45] Like, how do you describe the problem?
[20:48] I don't even know how to describe I
[20:49] know I've had many conversations where I I
[20:50] think everybody approaches a problem in that way,
[20:52] and then you're like, oh, no. No. This
[20:53] is definitely a physicist. Like, think of mindset.
[20:56] And it's, I think it's a very systematic
[20:59] way of breaking down problems and how you
[21:01] approach it that
[21:02] is there are so many things that is
[21:04] like, in biology, there are so many things
[21:05] that we can't account for or we don't
[21:07] know. And I know you kind of always
[21:08] think of the spherical cow of, like, okay.
[21:10] How close can we get? What do we
[21:11] need to take into our model? And having
[21:13] that approach,
[21:14] really helps us understand,
[21:16] you know, we're not gonna be able to
[21:17] take into account everything
[21:19] for the human body and where can we
[21:20] make those assumptions, but how do we do
[21:22] that in an appropriate way? And I think
[21:23] you kind of learn a lot of that
[21:24] from the physics education.
[21:27] So we're talking at the Heidelberg Law Firm
[21:29] where you were initially a student. What do
[21:32] you know now that you wish you'd known
[21:34] back then? Not about the Heidelberg Law Firm,
[21:37] specifically, but just about what would have helped
[21:39] you in your career.
[21:40] Oh, that's interesting.
[21:43] Things have changed a lot. Yeah. It's like
[21:45] the things have changed so much, and there's
[21:46] been I I I think,
[21:48] I I was always really excited about,
[21:51] yeah, like, interdisciplinary
[21:52] research and that side at the time. And
[21:53] I think just seeing I don't know how
[21:55] comfortable I would have been going at like,
[21:57] you know, now we're doing a virtual reality
[21:59] work. We're doing the wearable work, which I
[22:00] had no experience with at that stage. So
[22:02] I think just kind of knowing
[22:04] to pay more attention to even, like, some
[22:06] of the experimental data and some of the
[22:08] talks on, you know, how do you how
[22:09] like, you know, what where is where are
[22:11] the drawbacks of these medical imaging platforms? And
[22:13] I like,
[22:14] I I I liked going to talks outside
[22:17] of my field, but really understanding, like, how
[22:19] useful it was to kind of embrace those
[22:20] talks that might not be you might not
[22:22] immediately see the connection to your research, but
[22:24] it's worth going to and and can really
[22:26] spark ideas. I think,
[22:28] yeah, knowing that would have been helpful. And
[22:30] even just being more comfortable, like, even like,
[22:32] the students here,
[22:33] I like, are so comfortable raising their hand
[22:35] and asking questions. I think I I was
[22:37] a lot shyer as the person at that
[22:39] stage. I don't I don't know if I
[22:41] I really ever like, I I would not
[22:42] have been comfortable asking that. So I think
[22:44] just, you know, knowing it's okay if you
[22:46] don't know the perfect phrase or how to
[22:48] do it or, like, you know, you don't
[22:49] have to think through it. It doesn't have
[22:50] to be perfect to ask the question. I
[22:51] think it would have been useful to to
[22:53] realize them.
[22:54] And thinking about your job at the moment,
[22:56] again, this is in the context of giving
[22:58] careers advice to people who are starting out
[23:01] in their careers,
[23:02] what do you like best and least about
[23:04] your your current work, your current job?
[23:07] Like, I love that I I get to
[23:09] work on interdisciplinary work. I get to work
[23:10] with incredibly smart people. The students are amazing.
[23:14] We're in an engineering building where the hospital
[23:15] is across the street, so it is easy
[23:17] to walk over and ask for help. And,
[23:19] you know, as a physicist, like, going in
[23:21] and saying, like, this is what we think
[23:22] would be really useful and having them be
[23:24] like sometimes they're like, you know, that's not
[23:26] implementable, and it's cute, but never gonna happen.
[23:29] So it's it's kind of fun where you
[23:31] know? And I think a lot of people
[23:33] complain about writing grants and things sort of
[23:34] at the same time. It's like, you know,
[23:35] by writing the grants and doing this, you
[23:36] get to spend most of your time kind
[23:37] of daydreaming about what could we do and
[23:39] what are you know, dreaming big and thinking
[23:41] about, you know, high risk projects and then
[23:43] trying to figure out how to get there.
[23:44] We now have a really big team, so
[23:46] it's you know, we have enough people to
[23:48] really try out.
[23:50] We have a couple of really complicated problems
[23:51] that you break down and you have, like,
[23:53] five different people working on different pieces of
[23:54] it at the same time, and you can
[23:55] kinda see how that's all gonna come together
[23:57] when they're done. And that's it's really fun
[23:59] to kinda put that puzzle together and see
[24:01] if we can make progress in each of
[24:02] these pieces.
[24:04] I I yeah. I I really like the
[24:05] just getting to brainstorm the research projects and
[24:07] get to see where we can go, and
[24:08] they they kind of afford you that opportunity
[24:10] there, which is is great. And, like, now
[24:13] we have a new center where we have
[24:14] we have a whole section on wearables. We
[24:16] have another section of virtual reality and one
[24:18] on AI and one on high performance computing.
[24:19] We have other researchers and other faculty members
[24:22] kind of bringing that experience in to all
[24:24] really help push this idea of the digital
[24:26] twin and kind of having it be bigger
[24:27] than my lab is a lot more fun
[24:29] where it's like, let's bring
[24:31] in other people who have, you know, whole
[24:32] new ideas and and can go and, like
[24:34] and there there's people who are just amazing
[24:36] on the wearable side that are adding whole
[24:37] new dimensions you we wouldn't have
[24:40] expected. So that's been really fun.
[24:42] I'd say on the the negative side is
[24:43] it is I joined this and kinda went
[24:45] into this direction because I really love the
[24:46] physics. I love the coding. I'm the person
[24:48] who will stay up all night coding, and
[24:50] I love it. And this is the and,
[24:52] like, I do not write lines of code
[24:53] as much anymore. I know some people today
[24:55] were saying they still code every day.
[24:57] I do more of you know, I get
[24:58] to see code in our in our project
[25:00] meetings.
[25:01] So I'm I'm still in touch with the
[25:03] code, but I'm not the person who gets
[25:04] to run it. And I think that that
[25:06] that part is a little, like, a little
[25:07] bit of a change, and and there are
[25:09] a lot of meetings. So let's say between
[25:11] those those two things are the kind of
[25:13] drawbacks.
[25:14] I think you're the first person I ever
[25:16] asked that question. You said they actually like
[25:18] writing Grant post, but I can totally see
[25:20] what you described as. How what they mean.
[25:22] Yeah.
[25:23] Amanda Randalls, thank you very much. It's been
[25:25] a pleasure to talk with you. Thanks so
[25:26] much.
[25:36] That was Amanda Randalls of Duke University
[25:39] in conversation
[25:41] with Physics World's Margaret Harris.
[25:43] Thanks to both of them for coming on
[25:45] the podcast.
[25:47] Physics World will be at the APS
[25:50] Global Physics Summit,
[25:52] which will take place on March
[25:56] in Denver, Colorado,
[25:58] and online.
[26:00] At the largest physics meeting in the world,
[26:03] you can join thousands of physicists,
[26:06] students, and policy leaders for a week of
[26:09] connection and collaboration.
[26:12] Explore cutting edge science shaping our shared future
[26:16] and be part of the global physics community
[26:20] driving innovation
[26:21] forward.
[26:22] Explore the meeting at summit.aps.org.
[26:27] And when you're in Denver,
[26:29] look out for us at IOP Publishing's
[26:32] booth at the conference exhibition.
[26:36] I'm afraid that's all the time we have
[26:37] for this week's podcast.
[26:39] I'm Hamish Johnston, and our producer is Fred
[26:43] Iles.
[26:44] Our theme music is called one three seven,
[26:47] and it was composed and performed
[26:49] by the physicist
[26:51] Philip Moriarty.
[26:53] We'll be back again next week.

Shows are listed from Apple Podcasts' charts and directory, and Podcast Index where it is used. Every show and episode belongs to its publisher and plays from the publisher's own feed; this site hosts none of them.