Children's hospital uses open-source AI to model young hearts in seconds

Every child born with a heart defect has a slightly different heart, yet surgeons often have to choose from off-the-shelf devices. Children's Hospital of Philadelphia, known as CHOP, says open-source AI now lets its team build an exact 3D model of a child's heart from existing scans in seconds, before anyone enters the operating room.[1]

The account comes from an NVIDIA blog post that quotes CHOP cardiologist Dr Matthew Jolley. The figures are the hospital's own and have not been independently audited. The tools behind them, MONAI and SlicerHeart, are free and openly licensed.[1][3][2]

About 1% of live births involve a congenital heart defect, and no two are alike. CHOP's cardiac modelling service takes CT scans, MRI and 3D ultrasound that a child's care team already has, and produces anatomically precise heart models. The service runs on MONAI, an open-source medical imaging framework cofounded by NVIDIA.[1]

Jolley's lab first built SlicerHeart, a free extension to the 3D Slicer imaging software, to model small and complex child hearts. Each model once needed hours of work at a workstation. The team then trained segmentation networks with MONAI Label and NVIDIA's Auto3DSeg, starting once it had 10 or 20 image and model pairs.[1][2]

CHOP says the AI output meets the quality a trained person would produce, in seconds instead of about four hours. For complex holes between the heart's lower chambers, known as ventricular septal defects, the hospital now models routinely before surgery. It expects about 200 modelled cases this year.[1]

The post describes one child who had two failed repair attempts because surgeons could not locate the defect. A 3D model made the anatomy clear and the next repair succeeded first time. More than 20 US children's hospitals now run cardiac modelling programmes, and Boston Children's Hospital uses modelling for roughly 500 operations a year.[1]

The next goal is to predict what a device will do inside a particular heart. CHOP is working with NVIDIA and the open-source community on biomechanics tools for Newton, an open physics engine built on NVIDIA's Warp framework. The aim is to cut device simulations that can take four hours, or a whole night, to near real time.[1]

That work is early. CHOP has begun using Warp and Newton features for closure devices that seal holes in children's hearts and hopes to extend them to replacement valves. A link to 3D digital twins and virtual reality is still in development. Jolley says the patient group is too small and varied for normal commercial investment, which is why open source matters.[1]

Surgeons at the National Heart Institute, IJN, wrote in 2024 that Malaysia has about 500,000 births a year and roughly 5,000 babies with congenital heart disease. Ministry of Health data put the number needing surgery at 2,000 to 3,000 a year. IJN performs 800 to 900 of these operations, and eight government hospitals add 250 to 470.[4]

The same paper says elective cases at IJN wait at least a year, and that interventional cardiologists close some holes with devices to free operating theatre time. Faster planning tools could help in principle, but no Malaysian hospital is named in CHOP's work, and local use would need clinical validation.[4][5]