CHOP Cuts Pediatric Congenital Heart Disease Modeling Time from 4 Hours to Seconds Using Open-Source NVIDIA AI

Children’s Hospital of Philadelphia (CHOP) has reduced the time required to generate 3D heart models for pediatric patients with congenital heart disease from 4 hours to seconds by utilizing the open-source NVIDIA AI framework MONAI. This significantly accelerates clinical decision-making needed to select customized devices and plan surgeries for patients with complex anatomical structures.

Technical Innovation and Efficiency

Previously, it took skilled researchers approximately 4 hours to process CT, MRI, and 3D ultrasound data to create anatomically accurate heart models. CHOP introduced a method of training segmentation networks with image-model pairs by combining MONAI Label and NVIDIA Auto3DSeg. With only 10–20 training pairs, expert-level quality can now be achieved in seconds.

Clinical Impact and Adoption

This technology contributed to making pre-surgical modeling for complex ventricular septal defect (VSD) surgeries a standard of care. In two past cases of failed surgeries, 3D modeling clarified the defect location, leading to success on the first attempt. Currently, over 20 pediatric hospitals in the US operate this program, with Boston Children’s Hospital utilizing it for approximately 500 surgeries annually.

Open-Source Ecosystem and Expansion

Congenital heart disease patients are rare and diverse, lacking traditional commercial development incentives. Dr. Matthew Jolley emphasized that open source breaks down barriers for such small and heterogeneous populations and enables collaboration through a national pediatric hospital consortium. Additionally, GPU-based physics simulations using NVIDIA Newton and Warp are performing near-real-time device placement predictions, while integration of NVIDIA Omniverse and VLM is being pursued to enable intuitive interaction in VR environments.