Children’s Hospital of Philadelphia (CHOP) uses open-source artificial intelligence tools to build accurate three-dimensional models of the hearts of children with congenital defects, in an effort to improve procedure planning and reduce reliance on devices designed according to general sizing. According to material published on the NVIDIA blog, the system can produce the model within seconds, whereas the process previously required about four hours of manual work by a specialized researcher.
The service relies on images already collected as part of patient care, such as computed tomography, magnetic resonance imaging and three-dimensional ultrasound. CHOP uses the open-source MONAI framework for artificial intelligence in medical imaging, together with SlicerHeart, an extension for the 3D Slicer software dedicated to visualizing, segmenting and analyzing three-dimensional medical images.
From Research to Clinical Use
A team led by cardiologist and researcher Matthew Jolley began developing modeling pipelines for children’s hearts after he joined CHOP in 2015. Using MONAI Label and NVIDIA’s Auto3DSeg implementation, the team trained automated segmentation networks on pairs of images and previous models. The team says the output achieves the same quality standard as that produced by a trained specialist, but in far less time.
The hospital routinely uses these models before some complex operations, including cases involving ventricular septal defects, which are holes between the heart’s two lower chambers. The material describes the case of a child who underwent two unsuccessful repair attempts before the three-dimensional model helped surgeons identify the defect’s location and complete the repair on the first attempt. The model also allows comparison of the fit of two different artificial valves for the patient’s heart and estimation of their effect on the heart walls.
The use is not limited to CHOP; the material indicates that more than 20 children’s hospitals in the United States operate cardiac modeling programs. At Boston Children’s Hospital, this approach supports more than half of heart operations, approximately 500 cases annually, while CHOP expects to reach about 200 modeling cases during the year mentioned.
From the Shape of the Heart to Device Behavior
The team is currently working to add physics simulations to test what might happen when a device is inserted into a specific heart. This effort relies on Newton, an open-source physics engine built on the NVIDIA Warp framework in Python and using graphics processing units. These tools are being connected with SlicerHeart to estimate tissue properties and how defect-closure devices deploy inside the patient’s heart.
The material says simulations that may take up to four hours, or an entire night when testing multiple configurations, can approach real-time performance using GPU acceleration. The practical goal is to enable physicians to compare different devices within the time needed to make a decision on the same day, rather than merely viewing a static anatomical model.
Why Does This News Matter?
The importance of this development lies in combining patient imaging data with scalable open-source tools, particularly in a field involving rare and diverse conditions that do not always attract traditional commercial investment. CHOP points to collaboration among researchers from Stanford and Boston Children’s, as well as the formation of a national coalition of children’s hospitals to build shared modeling infrastructure.
However, the reported speed does not mean that simulation has become an independent substitute for medical decision-making; the material describes tools being integrated into clinical workflows, while some applications, such as transcatheter valve simulation and virtual environments, remain under development. The findings cited appear in material published by NVIDIA, so generalizability and independent clinical validation require further monitoring before they can be considered a uniform standard of care.