Exo

The models here make findings rather than measurements, and they reach into the act of scanning itself, which is a step beyond the two handheld competitors already graded in this index.

Three things distinguish the portfolio. Depth: 14 cleared indicators reported by June 2025, more than either competitor in this cluster. Kind: the indicators detect conditions rather than measuring structures, covering heart failure with reduced ejection fraction, cardiac hypertrophy, cardiomyopathies, valvular disease, pneumonia and pulmonary embolism. Detecting pneumonia is a different claim from measuring a tendon, because the output is a finding a clinician would otherwise have had to make. And placement: SweepAI evaluates image quality continuously as the operator sweeps and signals when enough diagnostic information has been captured, so the model is not analysing an image after acquisition but participating in whether an acceptable image exists at all.

The company's own framing supports the reading. It describes itself as a medical imaging software and devices company, software first, and the workflow product is explicitly built to run on other manufacturers' hardware, which is not something a hardware business does.

What holds it below the top grade is that the silicon remains a genuine asset. The transducer technology is proprietary and the device would be a differentiated product without a single model attached. Strip the models and a competitive handheld ultrasound at 3,500 dollars remains.

Graded B, a step above Butterfly Network and Clarius in this index, where the models measure and guide rather than diagnose.

Ask what each indicator outputs and how it is meant to be acted on.

Advisory by design, with a marketing claim that works directly against the oversight the design assumes.

The mechanics are sound. Indicators surface on the device in real time, the clinician sees them alongside the image they were derived from, and the clinician decides. Nothing acts on a patient and nothing records a diagnosis without a person. The acquisition assistant is better still as an oversight design, because rather than silently accepting a poor image it tells the operator when enough diagnostic information has been captured, which converts an invisible failure mode into visible feedback.

The tension is the superiority claim. Stating that a detection model outperformed clinicians alone is a reasonable thing to publish if true, and it is also precisely the message most likely to produce automation bias in the clinician the design relies on to supervise it. A clinician who has been told the tool is better than they are has little reason to override it, and the cases where overriding matters most are the unusual ones where the model is least reliable. That risk is created by the claim rather than by the software, which makes it a communication decision the vendor controls.

The user population sharpens it further. Nurses and emergency medical technicians are named as intended users, and an operator without imaging training cannot meaningfully second guess a finding on the image.

What is not described is behaviour at the limits: what an indicator does with an inadequate or out of scope image, whether it declines rather than guessing, and what confidence information reaches the user.

Graded B.

Ask what an indicator does on an out of scope image and what confidence signal is displayed.

The portfolio is described item by item and no item has a number attached.

Disclosure of what exists is good. Individual indicators are named rather than bundled, covering specific conditions including reduced ejection fraction, cardiac hypertrophy, cardiomyopathies, valvular disease, pneumonia and pulmonary embolism, so a buyer knows exactly which capabilities are claimed. The acquisition assistant's mechanism is described concretely, evaluating image quality continuously and signalling when sufficient diagnostic information has been captured. The transducer technology is characterised specifically as a piezoelectric micromachined approach on silicon rather than left as advanced technology.

What is absent is every performance measure. No model card, no sensitivity or specificity for any detection indicator, no agreement figure for any measurement, no error analysis, no validation population, no retraining cadence and no drift monitoring account were located.

The superiority claim makes this the sharpest transparency gap in the cluster rather than merely a familiar one. Asserting that a model outperforms clinicians while publishing no comparative figure inverts the normal position: usually a vendor has no evidence to publish, whereas here a comparison was evidently run and the result is characterised without being shown. The data exists.

A version question sits alongside it, since indicators arrive on devices already in the field and no published version history for the models was located.

Graded C.

Ask for sensitivity and specificity per indicator, the validation populations, and the comparative data behind the superiority claim.

The chain is undisclosed, with one mitigating feature the competitors in this cluster lack.

A dedicated pass located no sub processor register, no named cloud or infrastructure provider, no third party component inventory and no manufacturing partner for the transducer silicon. Nothing describes whether any of the 14 cleared indicators were developed in house, built with clinical or academic partners, or licensed from elsewhere, and no external model provider is named or excluded.

Training data provenance is unstated for every indicator, which is the most consequential omission. Fourteen detection and measurement models were built on imaging corpora that came from somewhere, and neither the sources nor the terms under which the images may be used are described. A competitor in this cluster publishes at least the corpus size for its obstetric model, so this is a disclosure norm that exists in the category and has not been met here.

The device agnostic workflow product adds a dependency question of its own, since integrating with nearly any point of care ultrasound device implies interfaces to other manufacturers' systems, none of which are named, and a customer cannot assess what happens when one of those relationships changes.

The mitigating feature is the trust centre. Unlike the other vendors in this cluster, this company operates a documented route through which a buyer can request security and compliance documentation, and a sub processor register is the kind of artefact such a portal often holds. That makes the information plausibly obtainable rather than absent, which is a meaningfully better position even though nothing was located publicly.

Graded D on what is published.

Ask the trust centre for the sub processor register, and ask which models were developed externally and what they were trained on.

The strongest comparative claim in this category, and the numbers behind it are not published.

The claim is the notable item. In announcing clearance for detection of pleural effusion and consolidation or atelectasis in June 2025, the company stated the tool showed superiority to clinicians alone. That is a comparative effectiveness statement rather than a performance statistic, and it is the sort of finding a regulator would have examined, which distinguishes it from marketing language. It is also unverifiable from public material: no accuracy figure, sensitivity, specificity, effect size, comparator population or study citation was located.

That gap matters more than usual because of what the claim implies. Superiority to clinicians is exactly the finding that would justify a clinician deferring to the tool, and a buyer cannot assess it without knowing which clinicians, at what experience level, on what case mix, and by how much. An emergency physician and a resident are different comparators, and the difference decides whether the claim describes a genuine advance or an easy baseline.

Other evidence is solid. Fourteen cleared indicators represent fourteen separate regulatory examinations. Wellstar MCG Health implemented the workflow product across its health system, which is a named institutional deployment. A named vice president of clinical affairs who is a recognised point of care ultrasound specialist gives the clinical programme identifiable ownership.

What is absent is outcome evidence: no study showing changed diagnosis, treatment or patient outcome was located, and no peer reviewed publication surfaced.

Ask for the study behind the superiority claim, including comparator, case mix and effect size.

Strong security governance around the data, and nothing about what the models are permitted to learn from it.

The custodial half is well evidenced and is graded on the security axis rather than repeated here: audited certifications, a public trust surface, and a stated boundary of responsibility covering the acquisition endpoint.

The stewardship half is unaddressed. Nothing located states whether clinical images acquired by customers are used to train, tune or evaluate the company's models, whether any such use is confined within a customer boundary, or whether a customer may decline. That question carries more weight here than for the competitors in this cluster, for a specific reason: this company ships more cleared models than either of them and continues to add them, so its appetite for imaging data is correspondingly larger, and its customers are generating exactly the data those models need.

The workflow product compounds it. Being device agnostic, it accumulates studies acquired on other manufacturers' probes as well as its own, which is a broader and more diverse corpus than any single device vendor's estate. Nothing describes whether that data is walled off from model development, and a customer scanning on a competitor's hardware may not realise the question applies to them.

No statement about training data provenance for any of the 14 indicators was located, which contrasts with a competitor in this cluster that publishes at least the corpus size for its obstetric model.

Graded C.

Ask whether customer images train or evaluate any model, whether the workflow product's data is separated from model development, and what a customer can decline.

Externally audited controls, a public trust surface, and an unusually honest statement about where the vendor's responsibility ends.

The evidence is substantial. The workflow product holds risk based two year certification under the healthcare assurance framework and completed an audited report at the operating effectiveness level, with information security management certification named alongside on a public trust centre. The framework in question is built on the federal health privacy and breach notification statutes, so certification against it is a direct mapping onto the regulatory obligations a covered entity cares about, examined by a third party rather than asserted.

The shared responsibility statement deserves particular credit. The company states plainly that the customer is responsible for the security of its own user equipment, meaning the phones and tablets the device pairs with, and that protecting those endpoints falls to the user. That is the honest position for any product acquiring clinical images on personal hardware, and most vendors in this category leave the boundary undefined rather than naming it. A buyer knowing where the vendor's obligation stops can plan for the gap; a buyer who assumes it does not stop cannot.

What is missing is the agreement itself. No template business associate agreement, breach notification window, liability cap position, audit rights statement or data return provision was located, though the trust centre provides a route to request documentation that most competitors do not offer.

Graded B.

Ask for the template agreement and breach notification window through the trust centre, and for the certification scope across device and workflow product.

Three externally audited credentials, a public trust surface, and continuity validated through the same audits.

The credential set is broad and correctly chosen. Risk based two year certification under the healthcare assurance framework is the demanding tier and maps onto the federal health privacy and breach statutes. An audited report at the operating effectiveness level tests whether controls actually operated across a period rather than whether they were designed correctly at a moment, which is the report health system procurement requests. Information security management certification covers the management system around both. Holding all three is rare in this index, and the security programme has a named executive owner.

The trust centre is what lifts this to the top grade. A buyer can reach the security documentation, understand what each framework covers, and request the audited report through a self service route rather than negotiating for it during a sales process. Publication rather than gating is the whole point of this axis, and only one other record in this index meets it as well.

Two further disclosures are worth noting because they are voluntary. Business continuity plans are stated to be validated through those same external audits. And the shared responsibility boundary is named, with the customer responsible for the security of the phones and tablets the device pairs with.

The one gap is currency and scope. The certifications were announced in 2023, scoped to the workflow product, and no coverage periods or renewal dates were located, so a reader cannot confirm the current standing or whether the device itself is in scope.

Ask for current coverage dates and whether the device and its models fall within the certification scope.

Fourteen cleared indicators, including a first of its kind detection clearance, on top of a cleared device.

The device pathway was established early, with clearance in 2021 and imaging modes and indications added in 2022. What earns the grade is what followed. By June 2025 the company reported 14 cleared artificial intelligence indicators embedded in the product, and cleared what it described as the first ever clearance for artificial intelligence detecting pleural effusion and consolidation or atelectasis.

The kind of clearance matters as much as the count. Most cleared ultrasound artificial intelligence in this index automates measurement, where the regulatory question is agreement with a human measuring the same thing. Detection clearances are harder, because the regulator is examining whether software correctly identifies the presence of a pathological finding, with the false negative consequences that implies. Obtaining a first of its kind detection clearance means there was no predicate to follow.

The acquisition assistant is separately cleared for cardiac and lung, which is worth noting because guidance software of that kind is frequently positioned outside device regulation as a usability feature.

Graded A, the third in this handheld ultrasound cluster. All three vendors in this category hold the strongest regulatory positions in the index, which reflects that imaging is a domain where the device pathway cannot be avoided, and all three have chosen to clear their models rather than only their hardware.

Ask for the clearance numbers and intended use statements for the 14 indicators, and which are cleared for which patient populations.

The company makes a bias claim, and it is about a different kind of bias than the one this axis measures.

The acquisition assistant is described as reducing operator dependence and bias, ensuring consistent and accurate imaging. Read carefully, that is a claim about variation between operators: two clinicians scanning the same patient should obtain comparable images. It is a real and worthwhile property. It is not a claim that the models perform equivalently across patients, and the wording invites the two to be conflated.

The distinction is the whole issue. Reducing operator variability standardises the input. It says nothing about whether a pneumonia detector performs equally well on a large patient and a small one, on a smoker's lungs and a child's, or across the populations the company explicitly targets in rural and under resourced settings. Ultrasound image quality varies substantially with body habitus, and a detection model trained predominantly on one distribution will underperform on another regardless of how consistent the sweep was.

The stakes follow from the indications. Pneumonia, pleural effusion and heart failure detection drive admission, antibiotic and diuretic decisions. A model that misses findings more often in one group produces silent undertreatment in that group, and because the intended users include nurses and pre hospital staff, there is often no expert reader to catch it.

A dedicated pass located no subgroup performance data, no training population description, no fairness review and no model documentation for any of the 14 indicators.

Ask what populations the detection models were trained and validated on, and how sensitivity varies by body habitus and patient group.

Recourse is undefined, and this record carries the sharpest version of the problem in the cluster because the models make findings rather than measurements.

A dedicated pass located no indemnification position, no warranty covering model output, no accuracy guarantee, no service credit regime and no described route to dispute a result.

The exposure follows from what the indicators do. A missed pleural effusion, an undetected consolidation or a wrongly reassuring ejection fraction estimate is a diagnostic failure, not a measurement discrepancy. In an emergency setting those findings drive whether a patient is admitted, given antibiotics, diuresed or sent home, and a false negative sends a patient home. The harm surfaces later, elsewhere, and traces back to a decision the clinician made with the tool's output in front of them.

The superiority claim complicates the allocation rather than clarifying it. If a vendor states its model outperforms clinicians alone, and a clinician consequently defers to it, the question of who is responsible for the resulting error is not obvious, and nothing published addresses it. A clinician who overrides a tool marketed as better than they are is exposed if they are wrong, and a clinician who defers is exposed if the tool is wrong.

The intended user population makes this concrete. Nurses and pre hospital staff are named users, working without an expert reader available, so the practical reviewer of a model finding may be the person least able to challenge it.

Graded D.

Ask what liability attaches to a false negative from a cleared detection indicator, and how responsibility is allocated when a clinician defers to a tool marketed as outperforming clinicians.

The workflow product is deliberately not tied to the company's own hardware, which is the strongest interoperability position in this cluster.

The workflow platform is described as integrating with nearly any point of care ultrasound device and with common hospital record and imaging archive systems, reachable from a phone, tablet or browser, and handling documentation, billing, quality assurance and education. Building software that works with competitors' probes is a genuine architectural and commercial choice, and it inverts the pattern of the other two handheld vendors in this index, whose workflow capability serves their own devices and whose imaging interchange is either sold separately or bundled into a membership.

The practical consequence is real. Health systems accumulate mixed probe estates over years, and the problem they actually have is that examinations acquired on several manufacturers' devices never reach the record consistently. A vendor neutral layer addresses that directly, and a named health system implemented it across its organisation.

The billing function is the part that closes the loop, because an examination that is documented but not billable is a cost rather than a service line, and this is the aspect of point of care ultrasound programmes that most often fails.

What holds it below the top grade is that the breadth claims are unspecified. No list of supported third party devices was located, no record or archive systems are named, and nothing describes whether integration is bidirectional or what configuration effort a deployment requires.

Ask which third party devices and record systems are supported by name, and what a deployment involves.

Continuity is documented and externally validated, which almost nothing else in this index can say.

The distinguishing disclosure is that the company maintains a framework to limit the impact of disruptive events on its operations globally, and states that its business continuity plans are validated on a regular basis through its third party external audit certifications. Business continuity is the commitment vendors in this index most consistently omit; naming it and tying its validation to audits already being performed is a materially stronger answer than a service level assertion, because it means an external party checked that the plan is viable rather than that it exists.

Deployment shape is clear. The workflow product is reachable from phone, tablet or browser with cloud stored data, and the device pairs with a clinician's own equipment. The boundary of responsibility is stated explicitly, with the customer responsible for securing the phones and tablets involved.

What is not stated is where data physically sits. No infrastructure provider is named, no region selection is described and no residency position was located, which matters for a company describing global operations and targeting under resourced settings across jurisdictions.

Tenancy is likewise unaddressed, and no recovery time or recovery point objective was located, so the continuity framework is evidenced in principle without a published target.

Graded B.

Ask for recovery objectives, where data resides by market, and whether tenancy is dedicated.

A starting price and nothing about what follows it.

The device was launched at a published starting price of 3,500 dollars, which is real disclosure and more than most of this index offers. It also positions the product deliberately, since the competitors in this cluster sit at roughly 2,995 and 3,899 dollars, so a buyer can place all three.

Everything recurring is undisclosed. No subscription or membership structure was located, which is the omission that matters most in this category, because both competitors graded here charge an annual fee and the terms of that fee are what determine five year cost. Whether the models are included with the device, unlocked by subscription, or licensed per indicator is not stated anywhere located, and with 14 cleared indicators that is a substantial open question. Nothing describes what happens to capability if a fee lapses, which is precisely where the two competitors differ most sharply from each other.

The workflow product is a second undisclosed line. It is sold to health systems, is device agnostic, and was deployed across a named health system, so it plainly has enterprise pricing, and no rate card, unit of charge or range was located.

No warranty terms, return policy or accessory pricing were found, which the closest competitor publishes in full.

Graded C: a headline price is published, the structure behind it is not.

Ask whether the cleared indicators are included with the device or licensed separately, what recurring fees apply, what happens when they lapse, and how the workflow product is priced.

Coverage designed around who is holding the probe rather than around anatomy, which is a different organising principle from either competitor.

The stated users are physicians, nurses, emergency medical technicians and clinicians working in emergency, acute care, outpatient and home settings, with rural and under resourced environments called out explicitly. Naming emergency medical technicians is not incidental: it places the device in a pre hospital setting where the operator has no imaging training at all and the alternative is no imaging.

Clinical coverage follows the acute use case rather than attempting everything. Cardiac and lung dominate, which is the right pair for undifferentiated breathlessness, the most common diagnostic problem this form factor is used for, with bladder, hip and thyroid applications alongside. That is narrower than a six probe specialty range and deeper on the indications an emergency clinician meets most.

The workflow product extends coverage past the company's own hardware. Because it is described as integrating with nearly any point of care ultrasound device, an organisation running mixed probes can standardise documentation and billing across all of them, so the addressable setting is larger than the installed base of the device.

What holds it at B is the absence of detail behind the breadth claims. Home and pre hospital settings are named as targets rather than evidenced with deployments, and no list of supported third party devices was located.

Ask which third party devices the workflow product supports, and for deployment evidence in pre hospital and home settings.