Children’s Hospital of Philadelphia is using open-source artificial intelligence tools to generate three-dimensional models of pediatric hearts in seconds, NVIDIA reported on Tuesday. The hospital’s cardiac modeling service relies on MONAI, an open-source medical imaging framework cofounded by NVIDIA, to convert CT scans, MRIs, and 3D ultrasounds into anatomically detailed heart models.
The automated process replaces an earlier workflow that required four hours of manual work by a skilled researcher. Roughly 1% of all live births involve a congenital heart defect, creating varied anatomies that off-the-shelf surgical devices rarely fit without customized planning.
Clinical adoption
More than 20 children’s hospitals across the United States currently operate cardiac modeling programs. Boston Children’s Hospital models roughly 500 cases a year, covering more than half of its cardiac surgeries, while CHOP expects to reach about 200 modeled cases this year. CHOP also plans to expand these modeling tools to other medical specialties through its IDEA Lab, part of the Morgan Center for Research and Innovation.
Dr. Matthew Jolley, a cardiologist and researcher who joined CHOP in 2015, developed specialized pediatric modeling workflows by collaborating with the open-source community to build SlicerHeart, an extension of the 3D Slicer imaging software. His team trained segmentation networks using MONAI Label and NVIDIA’s Auto3DSeg on pairs of past patient scans and models. CHOP now routinely models complex ventricular septal defects prior to surgery. In one early case, surgeons successfully located and repaired a defect on the first attempt after two previous operations without 3D modeling had failed.
Physics simulation
GPU acceleration through Newton, an open-source physics engine built on the NVIDIA Warp Python framework, is now expanding the hospital’s capabilities from anatomical visualization to biomechanical simulation. Testing how a closure device deploys and interacts with cardiac tissue previously required up to four hours or an overnight compute run. Running those simulations on GPUs cuts that duration to near real time, enabling same-day clinical evaluations.
Clinicians at CHOP have started applying Warp and Newton simulations to closure devices that seal holes in children’s hearts, with plans to expand the method to transcatheter valves. Researchers are also developing a coupler connecting SlicerHeart to NVIDIA Omniverse digital twins powered by OpenUSD, designed to feed patient simulations into virtual reality environments equipped with embedded vision-language models.
Shared infrastructure
A national consortium of children’s hospitals is now forming to establish shared open-source modeling infrastructure. With about 2.4 million people in the United States living with congenital heart disease, researchers at Stanford and Boston Children’s contribute code alongside CHOP to SlicerHeart, including joint research simulating how heart valves interact with blood flow.
