You can connect with the author here or here, or at [email protected]
Early last year, I chatted with Dr. Bobby Yanagawa, the head of cardiac surgery at St.Micheal’s Hospital in Toronto. Fun fact, I was born there! That wasn’t the reason we called though. I wanted to figure out how surgeons at one of Toronto’s largest hospitals predict blood flow in their patients before actually performing surgery on them.
We started Radiel Health with the hypothesis that medicine will become incredibly personalized going forward. When I was in grade 3, I got a rhinoplasty to fix my deviated septum and it didn’t work. It wasn’t because the surgeon was terrible, it was simply because they weren’t used to operating on children my age. Most of their previous knowledge for this surgery came from operating on adults, not a 9 year old. What if there was a way to predict the future of a surgery before it happens? That’s the problem we’re aimed to solve last year, and we decided to start with a huge problem area in this space, predicting blood flow.
Of course, that’s not a small ask. Being able to simulate the complicated physics of the way our blood behaves in our body is hard. Computational Fluid Dynamics (CFD) has been used as the best way to simulate blood flow for a long time, but traditional CFD solvers like OpenFOAM require hours of compute time per patient geometry [Morris 2016], preventing fast feedback to clinicians in time-sensitive clinical applications [Liang 2020].
Every time your heart beats, blood rushes through your arteries and drags against the walls, like water flowing through a pipe. That friction is called wall shear stress, and it turns out to be a big deal. When blood flows smoothly, it keeps your artery walls healthy. But at spots where arteries branch or curve, blood can slow down and swirl around. Those low-friction zones? That's exactly where plaque starts to build up. If a surgeon, can see where these “danger” zones could occur before they commit to a surgery, we believe that will reduce the number of failed vascular surgeries.
Thanks to our amazing mentor Dr. Sean Peterson from UWaterloo’s Fluid Flow Group, we took an approach to this problem that had us building our solution from the simplest thing possible. So, when we actually inference on a real vascular structure, we can be assured the physics is sound.

Example of a Lid DrivenCavity simulation
The Lid Driven Cavity is the simplest problem in fluid dynamics, I’m sure most engineering students reading this have gone through this tutorial at least once when first learning Ansys. Effectively there’s a 2D box like above, and the top wall moves to a given direction. The flow inside the box is what we take a look at! Well, not exactly inside the box, we instead focused on the outsides of the box, at the walls. That’s where the wall in wall shear stress comes from. The idea is that the insides of a fluid domain influence the walls, so if we’re able to predict the walls with a high degree of accuracy, the insides are a natural byproduct of that.
Speaking about byproducts, shout out to my little brother’s laptop. It was used extensively to run all the CFD simulations to generate the data for our model. He did tell me the other day his laptop now has a 20 minute battery life. Huh, what a coincidence! (R.I.P Pranav’s laptop)