The Success of ‘Open-hearted’ Partnerships in the Cloud

The future is open — and it’s better together. At SUSE, we pride ourselves on our partnerships, and sometimes what we can achieve together surpasses even our greatest hopes. That’s what our award-winning, cloud-based, high-performance computing (HPC) partnership with UberCloud, Dassault Systèmes, and Google Cloud achieved, by enabling 3DT Holdings researchers to create an affordable, real-time heart surgery simulator for physicians to use when it matters most. This is an ongoing relationship with the Living Heart Project that we think is just the beginning of what this ground-breaking research can achieve — and the lives it can save.

Leveraging machine learning during heart surgery 

Let’s start with the impressive medicine, and then we’ll talk about the technology that’s enabling it.

“If we apply the power of realistic simulation to human modeling, we can revolutionize medical care.” That’s the goal of Dassault Systèmes’ The Living Heart Project, in collaboration with the U.S. Food and Drug Administration. This multinational partnership aims to give the medical field access to the same predictive medicine and machine learning simulations already prevalent in the automotive, aerospace, energy, and high-tech industries. This includes giving surgeons access to this game-changer from inside the operating theater. At exponentially greater speed at a reduced cost. The Living Heart Project looks to examine the use of heart simulation as a safer way to expedite new device approvals, increasing innovation in addressing cardiovascular diseases, which is the leading cause of death worldwide.

The job of our four heart valves is to direct blood flow through the heart. Heart valve surgery is an option to correct when a valve either doesn’t fully open — valve stenosis or narrowing — or when it doesn’t close properly — valve incompetence, mitral regurgitation, or a leaky flow. One treatment for the latter is an artificial valve replacement. An alternative to high-risk open-heart surgery has physicians going in through a femoral catheter to attach a MitraClip to the mitral valve to restore normal blood flow. 

While this procedure has an overall lower risk, unsurprisingly, there’s less visibility than an open-heart surgery. The mitral valve is already anatomically complex, but 29% of patients have what is described as complex valve anatomy; 40% of those patients require multiple interventions to reposition the clip. This means that MitraClip positioning is crucial to increase success rates. Predictive modeling and simulations can quickly assist physicians before and during this minimally invasive heart surgery. 

Applying research from The Living Heart Project toward improving patient outcomes from valve replacement surgery, 3DT Holdings is building a decision-support solution that will deliver real-time guidance to heart surgeons in the operating theater — when it matters most. They began by creating a data replica of this surgery — based on echocardiograms taken before and after this surgery that’s performed hundreds of thousands of times a year in the U.S. alone — and using it to train a machine-learning model.

The result? Machine learning that used to take six to ten hours now takes about two seconds. And it comes with greater than 95% accuracy and at a lower cost. That’s not just real-time decision-making, it’s life-saving decision-making.

The technology driving real-time medical decision making

In order to achieve this, 3DT Holdings ran more than 3,000 heart simulations using Dassault Systèmes’ living heart geometry with Abaqus to train multi-stage machine learning with XGBoost, Cubist, and the Feedforward neural network.

This began as a high-compute workload that had to be performed at a workstation. A time-consuming workload. And an expensive one at that. 

3DT Holdings decided to partner with UberCloud HPC platform to move the simulations from on-premises to a containerized environment on the Google Cloud Platform in the Azure Cloud. UberCloud recommended they use Kubernetes clusters running on compute-optimized, preemptible virtual machine (VM) instances, and SUSE Rancher to supervise all 3,000 Kubernetes clusters — one for each heart simulation.

Daniel Gruber, director of architecture at UberCloud, said that it’d be impossible to manually supervise 3,000 Kubernetes clusters, and that this project couldn’t have happened without SUSE Rancher’s Kubernetes automation, monitoring and management. 

Gruber explained, “The crucial element in the solution is SUSE Rancher. Preemptible VMs are a great fit for 3DT Holdings, reducing cloud HPC costs by around 80%. However, these instances do not offer guaranteed uptime and can only stay running for a maximum of 24 hours. To eliminate the risk of simulations being shut off prematurely, we had to run each simulation in its own Google Kubernetes Engine cluster and automatically restart it at the end of the process.”

This technical partnership allowed 3DT Holdings to maximize its cloud resource and license usage. In the end, training this life-saving machine learning model cost less than $20,000.

“Our work with UberCloud and SUSE is playing an important role in developing an innovative solution for the heart-valve disorder, one of the health care industry’s most pressing challenges,” said Yaghoub Dabiri, scientist at 3DT Holdings.

The next step is 3DT Holdings to build a production-ready version that they will submit to the FDA for approval, with a vision to expand the project to train on echocardiograms performed during surgery, which in real-time can predict strains and stresses. Surgeons should be able to adjust in the moment instead of in subsequent operations.

The Living Heart Project applied the approach outlined in the WhitePaper: Deploying Kubernetes-based HPC Clusters in a Multi-Cloud Environment.

Award-winning medical and technical innovation

We are proud that this project was awarded the Editors’ Choice of HPCWire’s Best Use of HPC in the Cloud (Use Case) 2021. This award showcases the boundary-pushing research that can only be achieved through high-performance computing. We are honored to have played our small role in the future of personalized medicine that has the potential to improve the quality of life for millions.