NVIDIA and Children's Hospital of Philadelphia Generate Digital Hearts for Children in Seconds
The Children's Hospital of Philadelphia has created anatomically precise heart models using existing CT, MRI, and 3D ultrasound images through the open-source medical imaging framework MONAI, co-developed with NVIDIA. A process that previously took skilled researchers about four hours can now be completed in seconds, making it feasible for routine clinical use. Dr. Matthew Jolly stated that modeling is essential to find the right device for unique children before entering the catheterization or operating room.
The team trained segmentation networks using MONAI Label and Auto3DSeg on past images and models, producing outputs that meet trained human standards. GPU simulations have reduced the time for device simulations from up to four hours or overnight for multiple scenarios to near real-time, facilitating same-day comparisons of different valves or occluders to see if they compress tissue. The hospital has begun using Warp and Newton to simulate occlusion devices and plans to apply this to transcatheter valves; they are also connecting SlicerHeart to the Omniverse digital twin.
Congenital heart defects account for about 1% of live births and vary significantly. The hospital performs approximately 150 to 200 modeling cases annually, with over twenty children's hospitals in the U.S. undertaking similar projects; Boston Children's Hospital covers more than half of heart surgeries with modeling, totaling around 500 cases. One child had two failed repairs, and the location of the defect was elusive until the 3D model clarified the structure, allowing for a successful subsequent surgery.
Commercially, this represents a small and fragmented demand: few patients and diverse anatomies make it challenging for companies to invest long-term under conventional product logic. Several children's hospitals and NVIDIA have made tools open-source, relying on clinical feedback for iteration. Artificial intelligence reduces the cost of single-instance modeling, and open-source methods enable more hospitals to continue the work.
In market mechanisms, this is a computational power company exchanging open-source for surgical planning positions. What is sold is the assumption of a pediatric device market that sells software by case, while what is bought is the adoption by hospital clusters and the duration of GPU simulations. Funding does not manifest as a blockbuster medical software order but as research frameworks, accelerator cards, and subsequent digital twin coupling; beneficiaries are pediatric heart centers that can incorporate simulations into preoperative discussions, while those under pressure still rely on trial-and-error selection during surgery.
Open-source reduces the time spent on modeling but does not replace the surgeon's judgment regarding image quality and indications.
Source: Public Information
ABAB AI Insight
NVIDIA has connected the physics engine used for training robots to pediatric heart occluders, utilizing the same GPU simulation across different scenarios. The pediatric congenital heart disease market is too small for closed-source companies to invest four years of labor for each unique heart; open-source transforms labeling and segmentation into infrastructure that hospitals can continue to use, with NVIDIA providing acceleration and Omniverse interfaces. Jolly's comments highlight the matching issue: finished devices for unique anatomies, with preoperative digital trials replacing intraoperative trial-and-error.
The capital path involves leading with research frameworks, spreading clinical networks, and then binding real-time simulations and digital twins into the next layer. The 500 cases per year in Boston indicate that the method has moved beyond demonstration and into surgical lists. Relying on models for positioning after two failures turns rare, non-scalable structures into computable geometries.
A similar path is the diffusion of MONAI in radiological segmentation and the secondary development of hospitals after 3D Slicer becomes a standard in research. Medical AI is in a control phase: general models are not valuable for chat, but compressing four hours of labor into seconds and open-sourcing it to twenty hospitals is.
This represents a technological substitution. The mechanism is that the cost structure of rare diseases is rewritten by computational power; a small market does not mean no one is doing it, but rather that it must be open-sourced to be completed. Whoever can reduce single-instance labor can bring dispersed demand back into serious treatment consideration.
ABAB News · Cognitive Law
Diseases that are too small for the market can only sustain tools through open-source.
Seconds for modeling means buying fewer intraoperative errors.
Unique anatomies require the most repeatable calculations.