AI news videos

Open Source AI Models A Child's Heart In Seconds

NVIDIA says Children's Hospital of Philadelphia is using open source AI to build anatomically precise models of children's hearts in seconds, with the goal of safer and more precise care for kids with congenital heart disease.

Watch on YouTube

Transcript

Sources

Every claim in this video comes from the top ranking coverage of this topic. The claims and where each one came from:

• PUNCTUATION NOTE. The source article uses em dashes throughout its prose. This script file forbids dash characters of any kind, so every quotation reproduced below renders the source's em dashes as commas. No word inside any quotation was added, removed or changed. SENTENCE BY SENTENCE AUDIT. Hook: "NVIDIA says a Philadelphia children's hospital now builds heart models in seconds. The work runs on open source AI." HOOK RE-CUT NOTE: the hook was re-cut from one sentence into two AFTER the first draft, for a purely structural reason and with no change of meaning. The hookCardWave2 segment type takes exactly two on-screen beats, which is the convention in all 235 published data files that use it, so a single-sentence hook would have had to be split mid-sentence on the card. The original single sentence read "NVIDIA says a Philadelphia children's hospital now builds heart models in seconds using open source AI"; the trailing clause "using open source AI" became the standalone second beat "The work runs on open source AI." No fact was added or removed, and the open source characterisation remains supported by the same line quoted next, which describes MONAI as "an open source medical imaging framework cofounded by NVIDIA". Supported by the article's opening line "Children's Hospital of Philadelphia is using open source AI tools to model children's hearts in seconds, with the goal of enabling safer, more precise care for kids with congenital heart disease", and by the body line "CHOP's cardiac modeling service, built on MONAI, an open source medical imaging framework cofounded by NVIDIA, takes the images a child's care team already has, such as CT scans, MRI and 3D ultrasound, and produces anatomically precise heart models in just seconds." The institution is narrated as "a Philadelphia children's hospital" rather than by its full name Children's Hospital of Philadelphia, and never as the acronym CHOP, because the full name costs five words against a twenty word hook ceiling and the acronym is not self explanatory to a viewer; the city plus the words children's hospital identify it unambiguously and the title candidates carry the full institution name. "Model children's hearts" became "builds heart models" so the sentence reads as plain speech and avoids any wordplay on hearts. The whole hook is prefixed with "NVIDIA says" because this is NVIDIA's own blog making a capability claim about a platform NVIDIA cofounded, and no independent measurement of the seconds figure exists anywhere in the source. The article's framing goal, "safer, more precise care", is deliberately NOT narrated anywhere in this Short, because it is stated in the source as an aim and not as a demonstrated outcome. Scene one, single sentence: "MONAI is an open source imaging framework NVIDIA cofounded, and the hospital says it turns four hours of work into seconds." The first clause is supported by "CHOP's cardiac modeling service, built on MONAI, an open source medical imaging framework cofounded by NVIDIA". The word "medical" was dropped from "medical imaging framework" for length; the context of the Short is already a hospital, so the compression loses no meaning and claims nothing extra. The verb is the article's own "cofounded", not "built" or "owns", because NVIDIA cofounded MONAI rather than owning it outright. The second clause is supported by "A workflow that once required four hours of work by a skilled researcher now completes fast enough for routine clinical use", and the phrase "turns four hours of work into seconds" pairs that four hour baseline with the article's own seconds figure from the same paragraph, "takes the images a child's care team already has, such as CT scans, MRI and 3D ultrasound, and produces anatomically precise heart models in just seconds." Both numbers are the source's and so is the pairing, because the source writes the four hour line as the direct contrast to the seconds line. The subject of the saving is the modeling service, not the imaging data: an earlier draft read "existing images replace four hours of work", which wrongly credited the time saving to the images the care team already has rather than to the MONAI based modeling, and it was corrected before recording. The qualifier "by a skilled researcher" was dropped for length, which understates the source, since the source says those four hours were a trained specialist's hours. The clause is attributed with "the hospital says" because the four hour baseline and the claim of clinical grade output are the hospital's own account of its own workflow, reported by NVIDIA, with no audit, no timing study and no independent measurement anywhere in the source. The modality list, CT scans, MRI and 3D ultrasound, is quoted here but not narrated: it is a three item list against a two item ceiling, and the glued token 3D ultrasound is garbled by the speech engine. Scene two, single sentence: "About one percent of live births involve a congenital heart defect and no two are alike, so doctors model which device fits." The first two clauses are supported in substance verbatim by "About 1% of all live births involve a congenital heart defect, and no two are alike." The digit and the percent sign were written out as "one percent" per the numbers as words rule. The word "all" was dropped from "all live births" for length; "live births" was kept because it is the source's denominator and dropping it would silently change the statistic. The closing clause is supported by "A child born with a hole between the lower chambers of their heart, or a leaking valve in the single pumping chamber keeping them alive, needs care that fits their exact anatomy. Historically, the devices surgeons reach for were almost never designed with that specific child in mind", by the quoted line from Dr. Matthew Jolley, "You've got a one-of-a-kind kid and an off-the-shelf device. Our job is to find what fits, and modeling lets us do that before anyone goes into the cath lab or operating room", and by the figure caption "Using MONAI, CHOP can compare how two different artificial valves will fit in a patient's heart to improve surgery planning." Jolley is not named in the narration and none of his quoted wording is spoken: "one-of-a-kind" and "off-the-shelf" are hyphenated compounds that the speech engine garbles, and the second is also close to idiom. The narration states the plain fact underneath his quote, that doctors model the fit of a device, and it claims no outcome from doing so. The phrase "before anyone goes into the cath lab or operating room" was dropped for length, so the narration does not assert when the modeling happens; that omission only weakens the claim. The connective "so" carries the source's own causal framing, in which the variability of the anatomy is the stated reason the modeling exists. Scene three, single sentence: "NVIDIA says more than twenty children's hospitals model hearts and Boston Children's models over half its heart surgeries." Supported by "More than 20 children's hospitals across the U.S. now run cardiac modeling programs. At Boston Children's Hospital, modeling supports more than half of all cardiac surgeries, roughly 500 cases a year." The digits 20 and the phrase more than were written as "more than twenty" per the numbers as words rule. The qualifier "across the U.S." was dropped for length; this makes the geography vaguer but does not inflate the count, and the count itself is unchanged. "Run cardiac modeling programs" was compressed to "model hearts" for length. The source's "modeling supports more than half of all cardiac surgeries" was rendered as "Boston Children's models over half its heart surgeries", which keeps the same proportion and keeps the surgeries scoped to cardiac surgeries by saying "heart surgeries"; the scoping word was retained on purpose, because dropping it would wrongly imply more than half of every surgery that hospital performs. Word order makes the hospital the actor rather than the modeling, which is a readability change, not a strengthening. "Roughly 500 cases a year" and CHOP's own "about 200 modeled cases this year" were both cut for the word budget and are listed under deliberate omissions; both are additional detail in the same direction and neither contradicts what is narrated. The sentence is attributed to NVIDIA because these adoption figures appear only in NVIDIA's post and carry no citation, no registry and no third party count. Scene four, single sentence: "NVIDIA says Warp and Newton could cut device simulation from four hours to nearly real time, but that work has only begun." The first clause is supported by "Newton is an open source physics engine built on the NVIDIA Warp Python framework that runs physics simulations on GPUs", and by "With GPU acceleration, they can reduce the time needed for cardiac device simulation from up to four hours, or a full overnight run for multiple configurations, to near real time." The modal "could" was chosen deliberately in place of the source's indicative "can". That softening is intentional, because the surrounding paragraphs describe a capability being built rather than delivered: "CHOP is working with NVIDIA and the open source community to build biomechanics-focused simulation frameworks with Warp that can be brought into Newton", and "These frameworks, once integrated with 3D Slicer and SlicerHeart, can help doctors understand tissue material properties." The source's "from up to four hours" was shortened to "from four hours" for length, which slightly overstates the baseline, so the overnight run variant, which is the longer baseline in the same sentence, is omitted rather than used; taking the shorter of the two stated baselines keeps the claim conservative. "Near real time" was rendered as "nearly real time" because the hyphenated compound near-real-time is garbled by the speech engine. The closing clause, "but that work has only begun", is an in progress qualifier added by us and supported by "CHOP has begun to implement features built on Warp and Newton for the closure devices used to seal holes in children's hearts, and hopes to apply similar methods for simulations of transcatheter valves", by "The open source architecture of Warp and Newton is being connected with SlicerHeart, with the long-term goal of bringing real-time simulations into clinical workflows", and by the image caption "CHOP has started implementing features built on NVIDIA Warp and Newton to understand how a device will work in a patient's heart." That clause exists specifically to stop a viewer hearing scene four as a shipped product. The source's verbs are has begun, hopes, is being connected, long-term goal and in development; none of them describes a finished, deployed capability, and the narration says so plainly in the same breath as the speedup claim rather than in a separate sentence. Endcard: fixed channel boilerplate, carries no claim about the story. SENTENCE MERGE NOTE. Scenes one, two and four were each drafted as two spoken sentences and were deliberately merged into one sentence apiece after the first draft, joined with the connectives "and", "so" and "but", because full stops cost more spoken time than words on this voice and the lower sentence count buys duration headroom. The merge changed punctuation and connectives only. No fact was added, removed, softened or strengthened by it, and every clause still traces to the article lines quoted above. The one wording change made at the same time was in scene one, where "existing images replace four hours of work" became "it turns four hours of work into seconds", moving the credit for the time saving from the imaging data to the modeling service, as described in the scene one audit above. HONESTY CAVEAT. This is NVIDIA publishing on NVIDIA's own blog about open platforms that NVIDIA cofounded, funds and actively collaborates on, describing a hospital NVIDIA works with and quoting a doctor NVIDIA interviewed. It is a vendor telling a customer success story. It is not independent reporting, it is not a peer reviewed clinical result, and it is not a regulatory filing. The source contains no trial data, no cohort, no accuracy or error rate, no segmentation quality metric, no safety outcome, no complication rate, no comparison against a control workflow, and no independent evaluation of any kind. The single quality claim, that the machine learning output "meets the same quality standard a trained human would produce", is the team's own assessment reported without any measurement. The single patient story, the child with two failed repair attempts whose repair then succeeded on the first try, is one anecdote described by the hospital, with no follow up, no denominator and no way to attribute the outcome to the model; it was therefore deliberately left out of the narration entirely rather than narrated with a hedge. Accordingly every claim in this Short that goes beyond a plain fact about the world is attributed out loud: the hook and the seconds claim to NVIDIA, the four hours into seconds claim to the hospital, the adoption figures for twenty plus hospitals and Boston Children's to NVIDIA, and the Warp and Newton simulation speedup to NVIDIA with an explicit statement that the work has only begun. The one unattributed sentence is scene two, whose prevalence figure is background epidemiology rather than a claim about NVIDIA's products, and whose closing clause describes what doctors there do rather than asserting any performance or outcome. No marketing adjective from the source is repeated: "anatomically precise", "safer, more precise care", "industrial scale" and "profound" style framing are all absent from the narration. CORROBORATION STATUS. There is no corroboration for this story. The research step captured zero video transcripts: the box's yt-dlp build is too old and returned nothing for all five candidate videos, which is a tooling failure and carries no information about the story itself. The NVIDIA blog post, read from the local snapshot captured 2026-09-15T09:14:14Z and re-fetched live on the run date with a matching result, is therefore the single source, and every narrated sentence including the hook traces to a line quoted above. The timeline probes returned mostly unrelated videos and were not used, none were transcribed, none are cited, and nothing from their titles is narrated as fact. Because the story is same day, no timeline block was written into this script; adding one would age the story to the catalog tier. DELIBERATE OMISSIONS. Omitted for length, for speech engine safety, or because the claim is the vendor's framing: the name and title of Dr. Matthew Jolley and both of his quoted passages, including the phrase "machine learning has become just bread and butter", which is a banned idiom and was never a candidate for narration; the economics quote "It's too small a population to support traditional commercial development by normal economics" and "Open source defies traditional economics for small and heterogenous populations by allowing collaboration and progress without barriers"; the figure that about 2.4 million people in the U.S. live with congenital heart disease, and the related claim that the population has been too rare and too diverse to attract traditional device company investment; the whole decade long history section, meaning Jolley joining CHOP in 2015, 3D echocardiography coming online, the absence of tools for small complex anatomies, and the years when a single model meant a research assistant spending hours at a workstation; SlicerHeart and 3D Slicer by name, and the fact that SlicerHeart's tools are free to use and build on; MONAI Label and NVIDIA's Auto3DSeg implementation by name, both being glued brand and digit tokens that the speech engine garbles; the training detail about segmentation networks trained on pairs of prior images and models, and the ten to twenty image model pairs threshold; the modality list of CT scans, MRI and 3D ultrasound; the complex ventricular septal defect example and the statement that CHOP now models those routinely before surgery; the entire single patient anecdote of two failed repairs, surgeons unable to locate the defect, the model clarifying the anatomy and the repair succeeding on the first try; the line that cardiac modeling has moved from research to standard of care for cases like these; Boston Children's roughly 500 cases a year, and CHOP's expectation of about 200 modeled cases this year; the IDEA Lab and the Morgan Center for Research and Innovation and the plan to apply the same tools across multiple disciplines; the detail that Newton was originally intended for simulation based AI robot training; the overnight run baseline for multiple configurations; the closure devices used to seal holes in children's hearts, and the hope to apply similar methods to transcatheter valves; the scenario in which a clinician compares devices fast enough to inform a same day decision; the NVIDIA Omniverse and OpenUSD coupler that is in development, the digital twins framing, the virtual reality environments and the embedded vision language models, all of which are explicitly unshipped and all of which contain glued or hyphenated tokens; the figure caption about comparing two artificial valves with stress coloring on the cardiac walls, and the caption about simulating how heart valves interact with blood flow; the contributions of researchers at Stanford and Boston Children's and the national consortium of children's hospitals now forming; the closing paragraph about NVIDIA's investment giving small labs access to infrastructure maintained at industrial scale; and the unrelated related news links on the page about Perplexity Portable Computer, the IBC broadcast announcements, IFA 2026, and an NVIDIA acquisition item. Nothing omitted contradicts anything narrated. The omitted case counts, the omitted 2.4 million prevalence figure and the omitted patient anecdote all point the same direction as the narration; the omitted unshipped items, meaning Omniverse, OpenUSD, virtual reality and vision language models, would only reinforce scene four's statement that the simulation work has only begun. PROMPT INJECTION CHECK. The snapshot was read strictly as untrusted data and was checked for instructions. It contains no imperative addressed to a reader or to a model, no hidden directive, no "ignore previous instructions" pattern and no attempt to change behaviour. The imperative style text present is ordinary site furniture and sharing widgets: "Skip to content", "Share", "X", "Facebook", "LinkedIn", "Email", "Share on Mastodon", "Enter your Mastodon instance URL (optional)", "Share this Article", the friend's email address, your name, your email address and comments fields, "Send Email", "Email sent!", "Your browser does not support the video tag", the category and tag lists, and the closing marketing line "Learn how MONAI is advancing medical imaging AI." These are normal blog page elements and a product call to action, not injection. No injection attempt was found, and no text from the snapshot was treated as an instruction at any point.(NVIDIA (blogs.nvidia.com), first party corporate blog post titled "Heart of the Matter: How a Major Children's Hospital Uses Open Source NVIDIA AI for Cardiac Care", byline Isha Salian, page dateline September 15, 2026, published by NVIDIA about open source platforms NVIDIA cofounded and funds and about a hospital NVIDIA actively collaborates with, read from a local snapshot captured 2026-09-15T09:14:14Z and re-fetched live on the run date with a matching result)