Vscan Air

The artificial intelligence here is severable from the product, and the vendor's own support material says so: Vscan Air AI tools are optional and can be enabled if purchased, cardiac guidance appears in the tools menu only if bought, it is supported on the SL configuration only, and it is unavailable in guest mode. The US product page describes the CL and SL probes and does not mention the AI at all.

Without the software the product is a competent wireless dual probe handheld with B mode, M mode, color Doppler and pulsed wave Doppler, all working without a subscription. When enabled, the software does substantial work. Caption Guidance, the first AI authorized to guide the acquisition of medical images (DEN190040), gives turn by turn instruction toward standard echocardiographic views and captures automatically at diagnostic quality, and Caption Interpretation AutoEF (K210747) returns an ejection fraction from up to three views without manual border tracing. It remains a purchased option on one of two configurations, sold in the markets where it is authorized.

The oversight boundary is set in the authorization rather than only in policy. Caption Guidance's De Novo authorization (DEN190040) covers acquisition of echocardiograms for limited diagnostic use, so its output is expected to be read by someone qualified to read it, and the pivotal study was built that way, with nurses acquiring and cardiologists interpreting.

The design is directive: the guidance gives turn by turn prescriptive movements and then captures automatically once its own quality threshold is met, so the model decides when an image is good enough and the operator it was designed for is the one least equipped to disagree.

AutoEF returns a number without the border tracing a clinician would otherwise see, but K210747 describes the result shown with a confidence metric and a qualitative likelihood bin, views are labeled in the stored gallery, the report records which views it used, and the vendor states accuracy improves as views accumulate, which tells an operator that a single view result is weaker.

Nothing published sets a competency threshold, describes training beyond two remote sessions bundled with the subscription, or defines the escalation path when the automated number and the clinical picture disagree.

The approach and a version history are on the regulatory record, with training data only partly described. The two components are defined: Caption Guidance gives real time prescriptive movement instruction and automatic capture, and AutoEF calculates left ventricular ejection fraction from one or any combination of parasternal longitudinal, apical two chamber and apical four chamber views, with accuracy stated to increase as views are added.

K210747 describes AutoEF as machine learning, shows it as a new version of K200621 with its error measured against its predecessor (7.21 against 7.94 points RMSD), and says the update added 30 percent more training data from three ultrasound devices and two clinical sites. The supplementary material to the pivotal JAMA Cardiology publication describes the guidance model's training dataset, expert quality labeling and optimization.

What is not published is the architecture, the full training set size and composition, or how model versions reach devices through the subscription and whether a stored AutoEF report records the version that produced it; for a measurement compared serially in the same patient, silent model change is a reproducibility question.

The models are now the manufacturer's own cleared software, and their origin is traceable through the public record. Caption Health, founded in 2013 as Bay Labs, built the guidance and interpretation software, took Caption Guidance through De Novo (DEN190040, 2020) and AutoEF through 510(k) (K200621, K210747), published the pivotal trial in 2021, and was acquired by GE HealthCare in 2023, after which the software reached Vscan Air in April 2024.

Cleared models held in house cannot be quietly swapped. What is not disclosed is everything since the acquisition: whether the models have been retrained or revalidated under the new owner, on what data, and whether any third party or pretrained components sit inside them.

The platform supply chain is partly visible through the manufacturer disclosure statement program, which records whether a software bill of materials exists; no bill of materials for this product is published, and nothing names the cloud provider behind the storage tier.

The pivotal study is prospective, blinded and peer reviewed, and it predates this hardware. Published in JAMA Cardiology in 2021 from a registered trial, it paired eight nurses with no prior echocardiography experience, each scanning 30 patients across two hospitals with the guidance software, against sonographers scanning the same patients on the same hardware without it; five level three trained echocardiographers evaluated every acquisition independently and blind.

Acquisitions were of diagnostic quality for left ventricular size and function in 98.8 percent of patients, right ventricular size and function in 92.5 percent, and pericardial effusion in 98.8 percent, with at least 92.5 percent agreement between nurse and sonographer scans on the corresponding assessments. More than 90 percent of patients had cardiac abnormalities, and the primary endpoint was agreed with the FDA as part of the De Novo submission.

Two qualifications apply: several authors were employees of the company whose product was tested, so the study is company associated even though the readers were blinded and independent; and the trial ran on predecessor hardware, with the software since ported to Vscan Air under a new owner and no equivalent trial on the current platform. For AutoEF, K210747 reports a retrospective validation of 186 studies against the biplane method of disks (RMSD 7.21 points).

The vendor publishes a data rights clause in its product terms, and the clause is broad. It states that GE HealthCare may collect, prepare derivatives from and use non personal data related to its products, services and software services for purposes including training, demonstration, research and development and continuous product improvement, that it will own the resulting property rights, and that it will not sell the data or use it to identify the customer without consent.

The commitments not to sell and not to re identify are meaningful and are stated in the terms rather than in marketing. The problem is the undefined term the clause rests on. Non personal data is not defined, and in imaging the boundary is where the value sits: a de identified echocardiographic loop is the raw material for training an acquisition guidance model, and whether it falls inside the term decides whether a customer's studies feed the next model version.

No de identification standard is named, no opt out is described, and no distinction is drawn between telemetry and clinical images. Safety engineering, meaning guardrails and how a safety event is handled, is not published.

Device level privacy and security documentation is available, and no business associate position is published. GE HealthCare's product security portal describes a centralized repository for security updates and a cybersecurity risk management system, and supplies device privacy and security manuals and manufacturer disclosure statements (MDS2) to customers; whether a Vscan Air specific statement is published, rather than available on request, is not clear from the portal.

The product terms carry a data rights clause over non personal data. GE HealthCare does not state its status for the service. That question is live because the optional cloud storage and image sharing tier holds patient studies on vendor infrastructure, which is the relationship a business associate agreement governs.

GE HealthCare runs a product security program with an outside check behind it, and no independent certification is named. The product security portal describes a centralized repository for security updates and a cybersecurity risk management system, with sections for products and patches and for security notices, and supplies device privacy and security manuals and MDS2 statements to customers.

A coordinated vulnerability disclosure statement applies to all commercially available products and names a reporting channel, and GE HealthCare has published coordinated disclosures affecting its own ultrasound line, including a command injection and path traversal weakness reachable with physical access, crediting the external researchers and stating the residual risk decision openly. No audited certification such as ISO/IEC 27001, SOC 2 or IEC 62443-4-1 is named on the product security page, and it is unclear whether a Vscan Air specific disclosure statement is published.

The regulatory record is identifiable, and the AI's rests on a De Novo classification. FDA records list Caption Guidance under DEN190040 (7 February 2020, filed as Bay Labs, 21 CFR 892.2100, product code QJU), the first authorization of AI that guides the acquisition of medical images, which created a device classification rather than matching a predicate; a follow on 510(k), K201992, came in September 2020.

Caption Interpretation Automated Ejection Fraction Software is K200621 and then K210747 (19 January 2022, 892.2050, QIH), for adults, processing apical four chamber, apical two chamber or parasternal longitudinal views from ultrasound devices, computers or PACS. The Vscan Air probes carry their own clearances, including K202035 (2020), K231301 (August 2023) and K250087 (May 2025), 892.1550.

The Caption filings were made by Caption Health, which GE HealthCare acquired in 2023; the AI reached Vscan Air SL in April 2024. The vendor states that the AI is subject to government regulation and may not be available in all countries, and publishes an indications reference guide.

Subgroup consistency is reported for both AI components, as conclusions rather than full per group tables. The pivotal trial of Caption Guidance (JAMA Cardiology, 2021) reported that image quality results held across body mass index, sex and race, and BMI is the confounder that matters most in echocardiography because acoustic window quality varies with body habitus; more than 90 percent of enrolled patients had cardiac abnormalities on their full echocardiograms, so the result was obtained on abnormal hearts.

The 510(k) summary for Caption Interpretation AutoEF (K210747) balanced its test set for gender, ejection fraction and BMI and analyzed gender, BMI and multiple ultrasound manufacturers, and the summary gives no age, race or ejection fraction range breakdown. Three gaps remain. The trial is a 2021 cohort of 240 patients on predecessor hardware, and nothing published refreshes it for Vscan Air or for the model versions sold under the subscription.

Several authors were employees of the company whose product was tested. And there is no model card, no training population statement and no description of how a model update is revalidated against the same subgroups.

A published error rate for the automated ejection fraction is on the record, filed by the company GE HealthCare acquired, and nothing contractual stands behind it. The 510(k) summary for Caption Interpretation Automated Ejection Fraction Software (K210747, January 2022, filed by Caption Health, acquired by GE HealthCare in 2023) reports a retrospective validation on more than 186 studies with a biplane method of disks reference, balanced for gender, ejection fraction and BMI: a root mean square deviation of 7.21 ejection fraction points, against 7.94 for the previous version (K200621), with 1.09 percent of results off by more than 15 points and 0.55 percent by more than 20.

It is a test set result for adults, not performance in use, and it covers AutoEF only; Caption Guidance was authorized through De Novo (DEN190040), whose summary the FDA wrote. No warranty covering model output, indemnity, accuracy commitment or error route is published. The terms that are published cut the other way: a customer's failure to maintain software and handset compatibility for the AI is stated not to be grounds for non payment, and subscriptions renew automatically.

Interoperability is a purchase decision rather than a property of the device. Out of the box the probe streams to a validated phone or tablet and stores exams in the application; cloud storage, image sharing and the wider measurement toolset sit in an optional subscription tier, so without it the images are hard to move into an enterprise archive. AutoEF itself is cleared to process images from ultrasound devices, personal computers or DICOM compliant PACS (K210747).

The clinical record is untouched in either configuration: no HL7 interface, FHIR support or named EHR integration is published for the product, and an ejection fraction produced at the bedside reaches the archive only if the tier is bought and the archive is wired, not the note as a discrete coded result. GE HealthCare sells enterprise imaging and cardiology reporting products that could close the gap, as separate purchases rather than capabilities of Vscan Air.

The architecture is split, and only the local half is documented. The probe pairs wirelessly to a phone or tablet from a published compatibility list, the application runs on that device, and imaging and the AI execute against it, so a clinician can scan without a network; the compatibility list is a real operational disclosure because it tells a hospital which handsets will run the guidance software. The cloud half is where the questions sit.

An optional tier provides cloud storage and image sharing, so patient studies come to rest on vendor operated infrastructure, and nothing published states the hosting region, residency options, subprocessors, retention, or how a customer exports and deletes on exit. The AI is stated as possibly unavailable depending on regulatory approval in a given country, so a multinational health system can have different capability and different data paths by territory, with no published map.

A price is published, for the configuration without the AI. GE HealthCare's Vscan Air CL product page states a starting price of $4,999, with financing from $174 a month through Affirm, a three year warranty and a buy now link to the company's online shop. The terms state that subscriptions renew automatically for the same term, support material documents the AI as a purchased option enabled in the tools menu on the SL only, and availability is qualified by country.

The SL's price, each subscription tier and the price of the AI option, which is the feature that distinguishes this product, are not published alongside it.

The hardware covers the point of care range, and the AI covers one organ. The SL pairs a sector phased array with a linear transducer for cardiac, lung, vascular access and musculoskeletal work, aimed at cardiologists, intensivists and emergency physicians; the CL pairs a convex with a linear transducer for obstetric, gynecologic, abdominal and musculoskeletal imaging.

Named settings run across cardiology, critical care, emergency medicine, musculoskeletal, primary care and vascular access, and the device is verified for limited use outside professional healthcare facilities, an explicit statement of where it may be taken. The AI coverage is narrower and its restrictions stack: Caption AI is cardiac only, runs on the SL and not the CL, needs a handset from a validated list, is a purchased option, and may be unavailable in a given country. Its validation is in adults, from a two hospital trial and a retrospective AutoEF study.