Heart model production at Children’s Hospital of Philadelphia (CHOP) has gone from a four-hour job to a matter of seconds, according to a post NVIDIA published on 15 September 2026. The hospital’s cardiac modelling service takes the CT, MRI and 3D ultrasound scans a child’s care team already holds and turns them into an anatomically precise 3D heart model, using MONAI, the open source medical imaging framework NVIDIA helped to launch.
The NVIDIA post, written by Isha Salian under the headline “Heart of the Matter: How a Major Children’s Hospital Uses Open Source NVIDIA AI for Cardiac Care”, centres on Dr Matthew Jolley, a cardiologist and researcher at CHOP. “You’ve got a one-of-a-kind kid and an off-the-shelf device,” he told NVIDIA. “Our job is to find what fits — and modeling lets us do that before anyone goes into the cath lab or operating room.”
The story has three parts. The first is already routine: AI segmentation that produces a heart model fast enough for everyday clinical use. The second is under construction: physics simulations, built with NVIDIA Warp and the Newton engine, that would show how a closure device or valve behaves inside one particular child’s heart. The third is a plan rather than a product: a link to NVIDIA Omniverse and virtual reality so clinicians can explore the anatomy before they act.
This article explains each part, sets NVIDIA’s figures against CHOP’s own published research, and flags what the post does not say, including validation data and regulatory status. For the wider engineering picture behind tools like these, see our practical guide to medical imaging software development.
Table of contents
- What CHOP’s Heart Model Service Actually Does
- Why a One-of-a-Kind Child Needs a Personal Heart Model
- How AI Builds a Heart Model in Seconds
- From Heart Model to Operating Theatre: The Ventricular Septal Defect Case
- How Heart Model Programmes Spread to More Than 20 Children’s Hospitals
- Testing Devices Inside a Heart Model With Newton and Warp
- Omniverse, OpenUSD and Virtual Reality: Where the Heart Model Goes Next
- The Open Source Stack Behind the Heart Model Service
- Why Open Source Suits Congenital Heart Disease
- What NVIDIA’s Heart Model Account Leaves Out
- Lessons for Hospitals Building a Heart Model Programme
- Heart Model FAQs
- References
What CHOP's Heart Model Service Actually Does
At its simplest, the service converts scans into shapes. A child with congenital heart disease is imaged with computed tomography (CT), magnetic resonance imaging (MRI) or 3D echocardiography, the ultrasound technique that captures the heart in three dimensions. CHOP’s software then identifies each structure in those images and produces a heart model that surgeons and cardiologists can rotate, measure, print or examine in virtual reality.
The scans a care team already has
NVIDIA stresses that the service works from images the child’s care team already has. That matters for children, who may need many scans over their lives. The AI step does not require a new type of scan; it uses the CT, MRI and 3D ultrasound data that a paediatric heart centre collects for diagnosis anyway.
CHOP’s own published record shows why that detail counts. In a 2022 review of its clinical 3D modelling service, covering 2018 to 2020, the team found that 67% of cases were modelled retrospectively, from images already on file. Over time more requests arrived before the scan, so that image quality could be tuned for heart modelling.
What each imaging method brings to a heart model
Each modality has trade-offs, which Jolley set out in a February 2026 release from the Radiological Society of North America (RSNA) about his team’s MRI volume-rendering research. The release describes CT as giving excellent anatomical images while using ionising radiation and showing no blood flow. 3D ultrasound shows tissue and flow together but over a smaller field of view.
| Imaging method | Strength for a heart model | Limitation noted by CHOP researchers |
|---|---|---|
| CT | Excellent anatomical images | Uses ionising radiation; cannot show blood flow |
| MRI | High-quality flow images regardless of angle, with no radiation | Visual quality depends heavily on the underlying scan |
| 3D ultrasound | Real-time tissue and flow together, in a format cardiologists know well | Smaller field of view; flow accuracy depends on the beam angle |
MRI, by contrast, shows flow regardless of angle and without radiation, which Jolley called especially important for children who may need repeated imaging throughout their lives. His team presents its MRI techniques as a complement to ultrasound, not a replacement, and all three sources can feed a heart model.
From heart model to clinical decision
The output is not an end in itself. A heart model exists to plan an operation or a catheter procedure: where to make an incision, which patch or device will fit, and whether a repair that looks possible on a flat scan is possible in three dimensions. The RSNA release describes the same aim, seeing exactly where a defect sits so that its repair can be planned precisely.
Why a One-of-a-Kind Child Needs a Personal Heart Model
About 1% of live births involve a congenital heart defect, NVIDIA’s post says, and no two are alike. The US Centers for Disease Control and Prevention (CDC) gives a matching figure: heart defects affect nearly 1% of births each year, about 40,000 babies in the United States.
Off-the-shelf devices, one-of-a-kind anatomy
The problem Jolley describes is a mismatch of scale. Devices are designed for groups of patients large enough to justify a product line. A child born with a hole between the heart’s lower chambers, or a leaking valve in the single pumping chamber that keeps them alive, needs a repair that fits their exact anatomy. Historically, NVIDIA writes, the devices surgeons reach for were almost never designed with that specific child in mind.
A patient-specific heart model does not change the device catalogue. What it changes is the quality of the choice. Surgeons can check before the procedure whether a device of a given size and shape will seat where it needs to, or whether the anatomy calls for a different approach entirely.
How many children this affects
The CDC’s figures give a sense of the population. Of the roughly 40,000 US babies born with a heart defect each year, about one in four has a critical defect, the kind that typically needs surgery or another procedure in the first year of life. On the CDC’s own ratio, that is around 10,000 babies a year.
NVIDIA’s post adds a second number: about 2.4 million people in the US live with congenital heart disease. The CDC’s page says more than 2 million people are living with a heart defect, so the two are consistent, although NVIDIA does not name the source of its higher estimate.
Why rarity is the real constraint
Congenital heart disease is common as a category and rare as individual anatomy. Every child who needs a heart model is, in effect, a population of one. That is why heart modelling has grown up inside hospitals and research labs rather than device companies, and it is the thread that runs through the rest of NVIDIA’s account.
How AI Builds a Heart Model in Seconds
The slow step in modelling has always been segmentation: deciding, across every slice of a 3D scan, which parts are heart muscle, blood pool, valve tissue or vessel wall. It is a classic computer vision task, and for years CHOP did much of it by hand. NVIDIA’s post says a skilled research assistant used to spend about four hours at a workstation producing a single heart model.
CHOP’s 2022 review confirms the burden from the inside. “Image segmentation and CAD modeling is a time-consuming process requiring a highly skilled user,” the authors wrote. They also noted that the programme’s early development was largely the result of one person’s dedication during fellowship training.
MONAI Label and Auto3DSeg
The change came from machine learning. Using MONAI Label, a tool for AI-assisted annotation that plugs into the 3D Slicer imaging platform, and NVIDIA’s Auto3DSeg implementation, Jolley’s team trained segmentation networks on pairs of earlier scans and the heart model a human had built from each. According to NVIDIA, the output meets the same quality standard a trained human would produce, in seconds rather than hours.
The MONAI project lists Auto3DSeg as its reference implementation for automated segmentation pipelines, alongside models such as SwinUNETR and VISTA-3D. Automation is the point: it lets a clinical research team train new networks without hand-building every pipeline. The underlying method is the same family of image models explained in our guide to convolutional neural networks.
Ten or twenty examples to start
“Machine learning has become just bread and butter,” Jolley told NVIDIA. “As soon as we’ve made 10 or 20 image-model pairs, we train a model and start applying it.” That is a small number by the standards of general-purpose AI, which suggests each network is trained for a narrow task. The post does not say how many networks the lab now runs or how each was tested.
What “seconds” does and does not mean
Speed is the headline, but the useful measure is whether each heart model is right. NVIDIA says the output meets a human quality standard; it does not publish the accuracy metric, the test set or how often a clinician still corrects the AI. CHOP’s earlier paper describes digital models being verified for anatomic accuracy before clinical use, and a faster pipeline does not remove that step.
| Step | Before | With CHOP’s current tools | Status |
|---|---|---|---|
| Segmenting one heart model | About four hours by a skilled researcher | Seconds, with MONAI Label and Auto3DSeg | Routine |
| Starting a new model type | Every model built by hand | Train after 10 to 20 image-model pairs | Routine |
| Simulating one device deployment | Up to four hours | Near real time with GPU acceleration | Early implementation |
| Comparing several device configurations | A full overnight run | Fast enough for a same-day decision | Goal |
| Exploring anatomy in VR with AI queries | Not described | Omniverse coupler with embedded vision-language AI | In development |
From Heart Model to Operating Theatre: The Ventricular Septal Defect Case
The clearest clinical claim in NVIDIA’s post concerns ventricular septal defects (VSDs), holes in the wall between the heart’s two lower chambers. Complex VSDs, including multiple holes, can be hard to locate and repair. CHOP’s 2022 review lists repair of multiple VSDs among the most common reasons its team built a heart model in the first place.
Two failed repairs, then one that worked
NVIDIA describes one early case. A child had already undergone two failed repair attempts, with surgeons unable to locate the defect using traditional methods. A 3D heart model clarified the anatomy, and the next repair succeeded on the first try.
It is a single anecdote, and NVIDIA presents it as one: the case, in the post’s words, “made the value clear”. It is not evidence of outcomes across patients. CHOP’s 2022 review made the same distinction about its own data, stating that outcomes research on 3D visualisation was outside that paper’s scope.
Routine heart modelling before VSD surgery
The practical change is that modelling is no longer reserved for unusual cases. For complex VSDs, NVIDIA says, CHOP now builds a heart model routinely before surgery. Jolley describes cardiac modelling for these cases as having moved from research to standard of care.
What surgeons do with the model
CHOP’s earlier paper gives a concrete picture of how surgeons used models. With a flexible printed heart model, they could trial incision sites, check how well a VSD could be seen through the tricuspid valve and measure the geometry of abnormal vessels. Digital models let them design patches, baffles and conduits in computer-aided design software and view the planned repair in virtual reality.
The same paper shows which problems drove demand. A takeaway worth noting before the chart: VSD work was important, but complex biventricular repair was the biggest single reason for modelling.
The other indications included complex Fontan planning, ventricular assist device fit testing and reviews of earlier repairs. Percentages are rounded, so they sum to slightly more than 100.
How Heart Model Programmes Spread to More Than 20 Children's Hospitals
CHOP is not alone. NVIDIA reports that more than 20 children’s hospitals in the US now run cardiac modelling programmes. It singles out Boston Children’s Hospital, where modelling supports more than half of all cardiac surgeries, roughly 500 cases a year, and says CHOP expects to reach about 200 modelled cases in 2026.
CHOP’s heart model growth curve
Those numbers are easier to read against CHOP’s history. Its 2022 review counted 112 modelling cases between 2018 and 2020: 96 used clinically and 16 for education, with 127 individual models built for the clinical cases. Annual volume rose from 15 cases in 2018 to 47 in 2020, and modelling went from 8% to 27% of the hospital’s highest-risk surgeries.
On those figures, CHOP’s expected 2026 caseload is more than four times its 2020 volume:
The 2018 and 2020 figures come from CHOP’s paper; the 2026 and Boston figures come from NVIDIA’s post. The Boston number also implies a scale for that hospital’s surgical programme: if roughly 500 cases are more than half of its cardiac surgeries, the total is below about 1,000 a year. NVIDIA does not give the denominator, and it does not say whether Boston uses the same MONAI software.
From one lab to a shared toolkit
Part of the spread is shared software. SlicerHeart, the extension Jolley’s lab built for 3D Slicer, is free to use and extend, and NVIDIA says researchers at Stanford and Boston Children’s contribute tools alongside CHOP. A national consortium of children’s hospitals is now forming to build the next generation of shared heart modelling infrastructure, with open source as the connective tissue between institutions.
Beyond the heart: CHOP’s IDEA Lab
CHOP also wants to take the approach beyond cardiology. NVIDIA says the hospital aims to apply the same tools across multiple disciplines through its IDEA Lab, part of the Morgan Center for Research and Innovation. The post does not name the specialties involved or give a timeline.
Testing Devices Inside a Heart Model With Newton and Warp
A heart model shows what a child’s anatomy looks like. Jolley’s team wants to know what happens next: how a device will behave once it is deployed inside that anatomy, before any procedure begins. That is a physics problem, and it is where NVIDIA’s simulation software enters the story.
What Warp and Newton are
NVIDIA Warp is an open source Python framework that compiles Python functions into GPU code, so developers can write fast simulation software without low-level CUDA programming. It can also calculate gradients through a simulation, which helps when fitting physical parameters to observed data. Newton is an open source physics engine built on Warp and OpenUSD, co-developed by Disney Research, Google DeepMind and NVIDIA.
Newton was built for robots. The Linux Foundation took it on as a vendor-neutral project on 29 September 2025, and NVIDIA announced Newton 1.0 as generally available at its GTC conference in March 2026. Its Vertex Block Descent solver handles cables, cloth and volumetric deformable materials such as rubber parts.
From robot training to tissue behaviour
CHOP is working with NVIDIA and the open source community to build biomechanics-focused simulation frameworks with Warp that can be brought into Newton. Once they are integrated with 3D Slicer and SlicerHeart, NVIDIA says, the frameworks can help doctors understand the tissue material properties that determine how a device will deploy in a given patient.
That is a long way from robot locomotion and manipulation. Heart tissue is soft, moving and different in every child, and a simulation is only as good as the material properties fed into it. NVIDIA’s post describes the frameworks and their goal. It does not publish comparisons between simulated and actual device behaviour.
Four hours to near real time
The speed claim is specific. NVIDIA says GPU acceleration can cut the time needed for a cardiac device simulation from up to four hours, or a full overnight run for multiple configurations, to near real time. In practice, a clinician could compare how different devices fit a child’s anatomy and get results in time to inform a same-day decision.
At up to four hours a run, comparing even a few device options quickly becomes an overnight job:
The bars are simple multiplication of NVIDIA’s “up to four hours” figure. NVIDIA gives no number for the accelerated version beyond “near real time”, so it is not charted.
Closure devices first, valves next
CHOP has begun implementing features built on Warp and Newton for closure devices, the implants used to seal holes in children’s hearts. It hopes to apply similar methods to transcatheter valves, which are delivered by catheter rather than open surgery. The long-term goal, NVIDIA says, is real-time simulation inside clinical workflows. For now this is early implementation, not routine care.
Omniverse, OpenUSD and Virtual Reality: Where the Heart Model Goes Next
The most ambitious part of NVIDIA’s post is still in development. A coupler linking SlicerHeart with NVIDIA Omniverse digital twins, powered by OpenUSD, would let simulations built from patient images flow into virtual reality environments. There, clinicians could query and interact with a child’s cardiac anatomy before acting on it.
What OpenUSD adds
OpenUSD, short for Universal Scene Description, is a framework for describing 3D scenes that was created by Pixar. It is now developed through the Alliance for OpenUSD, whose founding members were Pixar, Adobe, Apple, Autodesk and NVIDIA. NVIDIA’s argument is that its open 3D interoperability lets different data sources and physics solvers be combined in one simulation built from a patient’s scans.
Asking questions of a virtual heart model
NVIDIA says the virtual reality environments would embed vision-language AI, systems that read images and text together, so clinicians can ask questions about the anatomy in front of them. The post gives no example questions, no timeline and no detail on how answers would be checked. It is best read as a direction of travel.
Virtual reality is already part of CHOP’s toolkit
Virtual reality itself is not new at CHOP. The 2022 review describes viewing models in VR through the SlicerVR module of 3D Slicer from July 2019, using a standard PC and a consumer VR headset, in a dedicated 3D imaging review suite. What would be new is interactive simulation and AI inside that space. For background, see our explainers on immersive technology and digital twin development.
The Open Source Stack Behind the Heart Model Service
Every layer NVIDIA describes is open source or built on an open standard, but the layers come from different places. Some are NVIDIA projects, some are academic, and some are run by foundations. The table below sets out who maintains each component and what it does in CHOP’s heart model workflow.
| Component | Role in the heart model workflow | Origin and governance | Licence |
|---|---|---|---|
| MONAI (Label, Core, Deploy, Auto3DSeg) | AI-assisted annotation, training and deployment of segmentation models | Launched by NVIDIA and King’s College London; maintained by NVIDIA, NIH, King’s, Mayo Clinic, MSKCC, Stanford, DKFZ and 30+ other institutions | Apache 2.0 |
| 3D Slicer | Visualisation, segmentation and analysis of 3D medical images | Community-developed research platform | Free and open source |
| SlicerHeart | Cardiac tools for 3D Slicer, including valve modelling and 3D echo import | Jolley’s lab and collaborators, with tools from Stanford and Boston Children’s | Free and open source |
| NVIDIA Warp | GPU-accelerated, differentiable simulation code in Python | NVIDIA | Apache 2.0 |
| Newton | Physics engine for device deployment simulation | Disney Research, Google DeepMind and NVIDIA; a Linux Foundation project | Apache 2.0 |
| OpenUSD | Shared 3D scene description for combining data and solvers | Created by Pixar; Alliance for OpenUSD | Open source |
| NVIDIA Omniverse | Digital twin platform for the planned VR coupler | NVIDIA | NVIDIA platform; not described as open source in the post |
MONAI’s London roots
MONAI has a UK connection that is easy to miss in NVIDIA’s description of it as a framework “cofounded by NVIDIA”. NVIDIA and King’s College London announced it in April 2020, and the project’s site now lists NVIDIA, the US National Institutes of Health and King’s College London among more than 30 maintaining institutions. It is licensed under Apache 2.0, which the project says permits use in commercial products and clinical pipelines.
Why the licence matters to a heart model team
Permissive licences are what let a hospital lab adapt tools rather than wait for a vendor. The 2022 SlicerHeart paper made the case directly: most commercial cardiac image analysis programs “cannot be significantly customized or extended by the end user to meet the needs of small or unique populations.” Open code removes that barrier, although it also moves responsibility for testing onto the hospital.
Where NVIDIA fits
NVIDIA’s role mixes maintainer and supplier. It helps run MONAI, Warp and Newton, and Warp’s GPU acceleration compiles to NVIDIA’s CUDA platform. Its post argues that open platforms give a children’s hospital lab access to tools “that would otherwise require a company-sized engineering team to build and sustain.” Both things can be true: the tools are genuinely open, and their fastest route runs on NVIDIA hardware.
Why Open Source Suits Congenital Heart Disease
The economic argument is the strongest part of NVIDIA’s post, and it rests on Jolley’s own words. “It’s too small a population to support traditional commercial development by normal economics,” he said. “But it’s such an important problem that between the research community and philanthropy, people are getting behind it.”
A market too small and too varied
NVIDIA frames congenital heart disease as a population too rare and too diverse to attract traditional device-company investment at the scale families need. No single company has built the tools Jolley’s team requires, the post says, and no single institution could build them alone. Open source, in Jolley’s words, “defies traditional economics for small and heterogenous populations by allowing collaboration and progress without barriers.”
The reimbursement gap in heart modelling
CHOP’s 2022 review points to a second economic problem. At the time, modelling was not directly reimbursed through traditional clinical mechanisms, and the programme’s facilities came from a competitive institutional grant. The authors noted four American Medical Association Category III procedure codes for 3D printed models and surgical guides, which track use but, until they become Category I codes, leave funding unsolved.
NVIDIA’s post does not address reimbursement at all. Faster AI segmentation lowers the labour cost of each heart model, which may matter more to a hospital’s finances than any change in billing, but no one has published that arithmetic for CHOP.
Research grants and consortia
What fills the gap, on Jolley’s account, is a mix of research funding, philanthropy and shared engineering. His lab’s profile lists National Institutes of Health grants for computer modelling of heart valves, and the forming national consortium would spread the cost of shared infrastructure across many hospitals rather than one.
What NVIDIA's Heart Model Account Leaves Out
NVIDIA’s post is a customer story on a vendor blog, and it reads like one: clear, positive and light on data. That does not make it wrong, and several of its figures line up with independent sources. But a hospital leader deciding whether to copy CHOP’s approach would need answers the post does not give.
| Claim in NVIDIA’s post | What we could check | Our reading |
|---|---|---|
| About 1% of live births involve a congenital heart defect | CDC: nearly 1% of births, about 40,000 US babies a year | Consistent |
| About 2.4 million people in the US live with congenital heart disease | CDC: more than 2 million people living with a heart defect | Consistent; NVIDIA’s source not named |
| A heart model takes seconds instead of about four hours | CHOP’s 2022 paper confirms manual segmentation was slow and skilled work | Plausible; no published timing data |
| AI output meets the quality of a trained human | No metric, test set or correction rate given | Unverified |
| More than 20 US children’s hospitals run modelling programmes | No list provided | Unverified |
| Boston Children’s models more than half its cardiac surgeries, about 500 a year | No Boston Children’s source cited in the post | Unverified |
| CHOP expects about 200 modelled cases in 2026 | CHOP’s paper: 15 cases in 2018, 47 in 2020 | Consistent with the growth trend |
| Device simulation from up to four hours to near real time | Frameworks in early implementation; no benchmark published | A goal, not yet measured publicly |
Validation and accuracy
The most important gap is accuracy. “Meets the same quality standard a trained human would produce” is a claim about results, and in medical imaging such claims are usually backed by overlap scores against expert segmentations, measured on patients the model has not seen and broken down by anatomy. None of that appears in the post, which is common for a vendor blog but matters to anyone hoping to replicate the heart model service.
Regulatory status
The post also says nothing about regulation. In the US, AI software that informs diagnosis or treatment can be a regulated medical device, and the Food and Drug Administration keeps a public list of AI-enabled devices authorised for marketing, last updated on 4 September 2026. CHOP’s 2022 paper said its segmentation then used FDA-approved software. NVIDIA does not say what regulatory pathway, if any, applies to the new AI heart model workflow.
Who checks the simulations
Simulation raises a harder version of the same question. A device simulation that gives a same-day answer is only useful if clinicians know how far to trust it for a particular child. The post describes material-property modelling as the aim, but not how the team will compare predictions with what actually happens in the catheter lab or operating theatre.
Lessons for Hospitals Building a Heart Model Programme
For hospitals outside the US, including in the UK and Europe, CHOP’s published experience is more useful than NVIDIA’s headline. The software is free to download, but the programme took a decade to build, and its own review is candid about what made the difference.
Start with image quality
CHOP learned in its first year that low-resolution scans, artefacts and images taken at an unsuitable point in the cardiac cycle increased segmentation time and reduced model quality. Setting guidelines for acceptable image quality improved both. AI does not remove that lesson: a network trained on good scans can struggle with poor ones, so imaging protocols come first.
Put the heart model where the surgeons are
The 2022 review credits a dedicated 3D imaging space, set up in 2019 close to the proceduralists, with increasing communication, collaboration and use of the tools. In 2018, 87% of requests came from the imaging team; by the most recent year, at least two-thirds came from the primary cardiac team, a sign that clinicians had begun to treat a heart model as part of planning.
Build on shared tools, and budget for people
Open source frameworks such as MONAI and SlicerHeart cut the cost of software, not the cost of expertise. Jolley’s lab combines cardiac surgeons, cardiac imagers, interventional cardiologists, engineers and computer scientists, according to its profile. Teams considering a similar route should plan for data governance, model testing and ML model development skills alongside their clinical champions.
It is also worth deciding early where patient images will be processed, a question we cover in private AI for healthcare, and how any model will be monitored once deployed, as discussed in our guide to machine learning in healthcare.
| Capability | Status in NVIDIA’s post | What to watch for |
|---|---|---|
| AI segmentation for each heart model | Routine at CHOP | Published accuracy and timing data |
| Modelling before complex VSD surgery | Standard of care at CHOP, per Jolley | Outcome studies across many patients |
| Closure device simulation with Warp and Newton | Implementation has begun | Predictions compared with real deployments |
| Transcatheter valve simulation | Hoped for | Any first results |
| SlicerHeart and Omniverse VR coupler | In development | Release and clinical testing |
| National consortium of children’s hospitals | Forming | Members and shared datasets |
| IDEA Lab use beyond cardiology | An aim | Named specialties |
Heart Model FAQs
What is a 3D heart model?
A 3D heart model is a digital, and sometimes printed, replica of one patient’s heart built from medical scans. Clinicians use it to understand complex anatomy and to plan surgery or catheter procedures. At CHOP, models are built from CT, MRI and 3D ultrasound images.
How does AI make a heart model faster?
AI speeds up segmentation, the step that separates heart structures in a scan. CHOP trained networks with MONAI Label and Auto3DSeg on earlier scans paired with human-built models. NVIDIA says the result meets human quality in seconds, compared with about four hours by hand.
Is CHOP’s heart model software free to use?
The underlying tools are open source. MONAI, NVIDIA Warp and Newton are released under the Apache 2.0 licence, and SlicerHeart and 3D Slicer are free to download. CHOP’s trained networks and clinical workflows are a different matter, and NVIDIA’s post does not say whether those are shared.
Is heart model simulation used on patients today?
Not routinely. NVIDIA says CHOP has begun implementing Warp and Newton features for closure devices and hopes to extend the approach to transcatheter valves. Real-time simulation inside clinical workflows is described as a long-term goal.
Does this help children outside the US?
Indirectly. The software is open, and MONAI was launched with King’s College London, so hospitals anywhere can use the same building blocks. The clinical results, caseload figures and consortium described by NVIDIA all relate to US children’s hospitals.
References
Heart of the Matter: How a Major Children’s Hospital Uses Open Source NVIDIA AI for Cardiac Care
SlicerHeart: An open-source computing platform for cardiac image analysis and modeling
SlicerHeart extension for 3D Slicer
3D Slicer image computing platform
3D MRI Technique Guides Precision Treatment of Kids’ Heart Conditions
Matthew A. Jolley, MD, CHOP Research Institute
About Congenital Heart Defects, CDC
London AI Centre and NVIDIA launch MONAI, a new AI framework for healthcare
Linux Foundation Announces Contribution of Newton by Disney Research, Google DeepMind and NVIDIA
Newton Adds Contact-Rich Manipulation and Locomotion Capabilities for Industrial Robotics
More AI coverage: explore Progressive Robot's AI Models, Tools & Releases hub — hands-on reviews, setup guides and benchmarks in one place.