CHOP uses open-source NVIDIA AI to model kids’ hearts in seconds

NVIDIA’s blog said Children’s Hospital of Philadelphia is using open-source AI tools — including MONAI, a medical-imaging toolkit NVIDIA helped start — to turn CT, MRI, and 3D ultrasound into anatomically precise models of children’s hearts in seconds. The hospital says that work once took about four hours, and it is also building simulations to test how devices fit before a procedure.

Kids born with heart defects often have anatomy no catalog device was made for. A same-day NVIDIA blog says a major children’s hospital now builds precise heart models in seconds instead of hours — and that more than 20 U.S. children’s hospitals run similar programs.

NVIDIA published “Heart of the Matter: How a Major Children’s Hospital Uses Open Source NVIDIA AI for Cardiac Care” on its company blog on 15 Sep 2026, by Isha Salian. The post describes a cardiac modeling service at Children’s Hospital of Philadelphia, or CHOP. That NVIDIA blog is the filing event. It is a company and partner case study, not a peer-reviewed trial paper. Clinical claims below are CHOP clinicians as quoted or summarized in that post.

The workflow, as NVIDIA and CHOP tell it, takes images a child’s care team already has — CT scans, MRI, and 3D ultrasound — and produces anatomically precise heart models in seconds. NVIDIA’s post says a workflow that once required about four hours of work by a skilled researcher now finishes fast enough for routine clinical use. File the time cut as NVIDIA/CHOP’s. This desk did not time a case.

About 1% of live births involve a congenital heart defect, the blog says. Congenital means present from birth. The same post says defects vary widely — no two are alike. File 1% and the variation as NVIDIA’s blog figures, attributed to the CHOP case.

Named voice: Dr. Matthew Jolley, a cardiologist and researcher at CHOP, quoted on the NVIDIA blog: “You’ve got a one-of-a-kind kid and an off-the-shelf device.” And: “Our job is to find what fits — and modeling lets us do that before anyone goes into the cath lab or operating room.” A cath lab is the room where doctors thread a catheter — a thin tube — through blood vessels. File the quotes as his, via NVIDIA. They are not an independent product review.

Tools named on the blog: SlicerHeart, an extension of 3D Slicer, the open software used to view and cut up 3D medical images; plus MONAI Label and NVIDIA Auto3DSeg, used to train segmentation networks on prior pairs of images and models. MONAI is an open medical-imaging AI toolkit cofounded by NVIDIA. Segmentation means teaching software to outline organs in a scan. File the tool names as NVIDIA/CHOP’s. How the networks are wired stays in Sources.

Clinical examples, still the NVIDIA blog’s CHOP anecdotes: the hospital now models complex ventricular septal defects — holes between the heart’s lower chambers — routinely before surgery. One early case, the post says, involved a child who had already had two failed repairs; the 3D model clarified the anatomy and the next repair succeeded. File that as a CHOP anecdote via NVIDIA — not an independent trial result, not a mortality claim, and not FDA clearance.

Scale figures, still NVIDIA’s blog: CHOP expects about 200 modeled cases this year. Boston Children’s Hospital is cited as using modeling for more than half of its cardiac surgeries — about 500 cases a year. More than 20 U.S. children’s hospitals now run cardiac modeling programs. File 200, 500, more-than-half, and 20-plus as NVIDIA’s blog figures. This desk did not audit those caseloads.

Simulation track, forward-looking: the blog says CHOP is building biomechanics simulations with Newton, an open-source physics engine that runs on NVIDIA Warp, aiming to cut device-fit runs that can take up to four hours — or overnight for several configurations — toward near-interactive results. File that aim as NVIDIA/CHOP’s. It is not a cleared product claim, and this desk did not run a simulation. Extra 3D-platform coupler talk stays in Sources.

About 2.4 million people in the U.S. live with congenital heart disease, the same blog says. File 2.4 million as NVIDIA’s figure. This desk did not recount a registry.

PRIMARY here: NVIDIA’s 15 Sep 2026 blog — Tier A company/partner record of what CHOP is deploying with NVIDIA’s open-source stack. The seconds-scale models, the about-four-hour baseline, the Jolley quotes, the SlicerHeart / MONAI / Auto3DSeg names, the ventricular-septal-defect and two-failed-repairs anecdotes, the 200 / 500 / 20-plus caseload figures, the Newton / Warp simulation aim, and the 2.4 million living-with-CHD figure are NVIDIA- or CHOP-attributed via that post. The same-day NVIDIA Newsroom index lists the blog as a link-out, not a second originating text. NOT claimed: FDA clearance, a peer-reviewed trial endpoint, a mortality reduction, that NVIDIA cures heart disease, independently timed model-build minutes, that this desk watched a case, a stock tip, or investment advice. Distinct from the already-filed curvebeam-ai-fda-ct-bmd-510k, inductive-bio-indy-chemistry, tandem-health-100m-series-b, and openai-glass-imaging-300m.