High-Performance Computing for the Living Heart Project Part - 2

Steven Levine, Dassault Systèmes , Founder and Executive Director of the Living Heart  Project

Wolfgang Gentzsch, President and Co-founder of Simr

In Part 1 of our series on High-Performance Computing for the Living Heart Project, we focused on the overview and current status of the Living Heart and Living Human Project by Dassault Systèmes. In this second part of the blog, we will delve into specific Living Heart and Brain projects conducted in the cloud, involving end-users from Stanford University, NIMHANS (National Institute of Mental Health and Neurosciences) in India, and healthcare device companies ADMEDES, ENMODES, and 3DT Holdings.

Below, we provide a brief summary of each project. More information is available in the webinar recording and slides (available upon request).

Project 197: Studying Drug-induced Arrhythmias of a Human Heart

Goal: Create a biventricular FEM model to study drug-induced arrhythmias of a human heart

• Electric waves in the heart turn chaotic, caused as side effect of drugs
• To predict response of the heart by measuring the effect of each drug
• Highly accurate personalized digital human heart models with 7M elements, 250M internal variables, 1M time steps
• Establish a unified foundation for cardiovascular in silico medicine
• 42 Abaqus simulations for 42 drugs each took 40h on 5-nodes (160-core)

Project 200: Simulation of Neuromodulation in Schizophrenia

• Modeling & simulation to improve clinical application of non-invasive transcranial electro-stimulation of the human brain in schizophrenia
• A novel ambulant personalized method for Deep Brain Stimulation
• Allows patient-specific interactive real-time treatment

Project 215: FSI Simulation of Artificial Aortic Heart Valves

• Assist surgeons by evaluating the clinically available devices in virtual native anatomy of the patient
• Solve complex multi-physics fluid-structure interaction (FSI) problem of native and prosthetic valves
• Understand dynamic behavior of the valve and its effect on hemodynamic
• Study different physics parameters (pressure, velocity, strain, stress) in the aortic valve region while applying the drugs by blocking different ionic currents in the cellular model

Project 216: Simulation of a Personalized Left Atrial Appendage Occluder (PLAAO) Device

• Virtually implant personalized braided nitinol stent
• Analyze loading of the device in different positions during the motion of the beating heart
• Reduce complications like leakage or migration of the device
• Design patient specific devices using Living Heart Model based on the patient anatomy

Project 222: Simulation of Cardiac Valve Disease with Machine Learning

Goal: Provide a real-time surgery simulator to heart surgeons for repairing cardiac valve leakage with a MitraClip device.

Machine Learning: 12h – 24h simulation => 2 - 3 secs prediction
Focusing on practical industrial usability, not supercomputers
Using just ‘cheap’ Cloud HPC

• Challenge: Repairing cardiac valve leakage (heart’s mitral valve leaflets don't close tightly), most common cardiac valvular disease, with 6+ million people in the U.S.
• Solution: MitraClip (MC), a small metal clip allows doctors with catheter-based surgery to repair the mitral valve, brings minimally invasive alternative to open-heart surgery.
• Approach: Compute: Google Cloud +Simr Platform, K8s Management: SUSE Rancher in Simr’s subscription, Abaqus container + DCV remote viz, Dassault’s LHP model.

• Results:
  

One Abaqus simulation on engineer’s on-prem workstation* = 12 - 20 hours

3,000 jobs on workstation = 1,500 days     

One Abaqus job on GCP on 1 compute node =  4 hours

On GCP 60 x 50 jobs in parallel = 12 days

For less than $20K =>  ML prediction with 95+ % accuracy = 2 secs

*Compute node: two Intel Xeon E5-2680 v4 processors, each with 14 cores

For more detailed information, please refer to the webinar recording (slides available upon request).