Children’s Hospital of Philadelphia is using machine learning to create patient-specific heart models from CT, MRI and 3D ultrasound scans in seconds. The same workflow previously required about four hours of work by a skilled researcher, making it difficult to use routinely before procedures.
The service is built on MONAI, an open-source medical-imaging framework co-founded by Nvidia, plus the SlicerHeart extension for 3D Slicer. The hospital trained segmentation networks with MONAI Label and Nvidia’s Auto3DSeg using earlier image-and-model pairs. Clinicians can inspect a child’s individual anatomy before entering a catheterization lab or operating room.
CHOP now routinely models complex ventricular septal defects. More than 20 US children’s hospitals operate cardiac-modeling programs; Boston Children’s uses modeling in more than half of its cardiac surgeries, about 500 cases annually, while CHOP expects around 200 modeled cases this year.
The team is also testing GPU-accelerated physics tools built with Nvidia Warp and Newton to simulate how closure devices or valves may behave inside a particular heart. That work aims to reduce simulations from hours to near real time, but it is still being integrated. The current clinical advance is faster anatomical modeling, not an autonomous diagnosis or treatment system.