Mindray TE Air

This is the thinnest artificial intelligence in the handheld cluster and the grade should be read as a boundary judgement rather than a criticism of the device.

What ships is recognition in service of setup and measurement. Scene recognition detects the anatomy under the probe and switches and optimises the preset between abdomen, thyroid and carotid. iTouch+ optimises the image in one action. PW Auto Calc traces a pulsed wave spectrum automatically and returns peak systolic and end diastolic velocities, time averaged maxima, resistance and pulsatility indices and heart rate without manual caliper work. Smart Bladder recognises three diameters automatically and computes a volume. Those are genuine models and they are why this record is admitted rather than screened out.

What is absent is any model that produces a clinical finding or a quantification a clinician would treat as a result in itself. There is no automated ejection fraction here, no B line counting, no inferior vena cava or velocity time integral tool. That set exists in the manufacturer's line and runs on the trolley and portable systems, reachable from this device only by docking the phased array probe into one, which makes it a capability of the cart rather than of the handheld.

The commercial framing confirms the reading. The product is sold on image quality, ruggedness, battery life, a three year warranty and the absence of subscriptions, and some smart tools are offered as optional one time upgrades rather than as the reason to buy. Graded D: models are present, they automate the work around the exam rather than any part of the exam's interpretation, and a buyer choosing this device is choosing an imaging instrument first.

The inherent autonomy risk is the lowest in this cluster, because nothing here interprets. Scene recognition changes a preset, one click optimisation adjusts an image, automatic spectrum tracing draws calipers on a Doppler waveform, and bladder volume returns a number from three recognised diameters. A clinician is reading the image throughout and every output is visible on screen next to the anatomy that produced it.

The grade is held at C because low risk is not the same as documented oversight, and one specific failure mode goes unaddressed. Automatic preset switching changes imaging parameters silently in response to what the model believes it is seeing. If the model misreads the anatomy, the operator is handed an image optimised for the wrong structure and has no obvious prompt that a decision was made on their behalf. Nothing published describes whether the switch is announced, whether it can be locked out for a given exam, or how an operator overrides it. The same question applies to automatic caliper placement on a Doppler spectrum, where the difference between a correct and an incorrect trace is a number a clinician may carry forward.

Nothing published sets competency expectations, describes training that accompanies the automated features, or defines what a user should do when an automated measurement disagrees with their own. The vendor does provide tutorial tools inside the application, which is a partial answer aimed at technique rather than at oversight.

Graded C: a design that keeps the clinician in the loop by construction, with no documented controls around the automation it does perform.

Hardware disclosure is specific and model disclosure is functional only, a pattern common in this cluster and unusually stark here because the hardware detail is so good.

On the instrument the manufacturer names its imaging platform, its second generation single crystal transducer technology and its matrix thin slice processing, publishes frequency ranges, scanning depth to 40 cm, ingress protection ratings, battery endurance and charge times, and explains the software based beam steering that lets a phased array probe simulate a convex footprint. The clearance summaries add imaging modes and probe architecture. A reader can form an accurate picture of what the device is.

On the models the disclosure stops at what each one does. Scene recognition is described as detecting anatomy and switching presets across three named regions, automatic spectrum tracing by the indices it returns, bladder volume by the three diameters it recognises. Nothing describes how any of them works, what data trained them, how often they change, or whether a result records the version that produced it.

One boundary deserves recording because the manufacturer's own marketing blurs it. The extensive smart tool set promoted across this ultrasound line, covering B line counting, velocity time integral, inferior vena cava, fetal heart rate and automatic ejection fraction, belongs to the trolley and portable systems rather than to this handheld. A buyer reading vendor material across the family could easily assume those tools come with this device. They do not, unless the phased array probe is docked into a system that has them.

Nothing upstream is disclosed. No third party model supplier is named, no statement establishes whether the recognition and measurement models are developed in house or licensed, no training corpus is described, and no pretrained or open source component is acknowledged.

This is the weakest supply chain disclosure in the handheld cluster and the contrast is sharp. Philips names the outside developer of its cardiac algorithm in a public clearance summary and discloses public co funding of its lung work. GE HealthCare's models have a fully traceable public history through an acquisition. Both are auditable by a buyer. Here a diligence team has no starting point.

One partial credit applies. The manufacturer states that software bills of material are among the artefacts it provides to customers, which is the mechanism through which third party and open source components would become visible. No bill of material for this product was located publicly, so the mechanism is offered and the artefact was not found, and a bill of material would in any case describe software dependencies rather than model provenance.

The hardware supply chain is comparatively visible, with the imaging platform, transducer crystal technology and processing named as the manufacturer's own, and manufacture is disclosed as taking place under the company's own quality systems. That is vertical integration rather than disclosure, and it does answer part of the question a buyer asks, which is whether the thing was assembled from components nobody can account for.

Graded D as an absence of disclosure, not as evidence of a problem.

The device is evidenced and the models are not, and here the split is wider than on any neighbouring record.

Independent work on the hardware is real and reasonably varied. A prospective comparative study assessed cardiac performance. A pilot study examined hepatic and renal vessels in 61 healthy volunteers across a wide age range using B mode and colour coded Doppler, rating images on a defined scale. A separate cross sectional comparison of six portable devices ranked this one highest for the apical four chamber view on criteria including endocardial definition, valve leaflet clarity, far field resolution and colour flow performance, and placed it among the top three for overall satisfaction on ease of use and image quality. For a value positioned device, being rated top of a field for the hardest cardiac view is a meaningful result.

Two qualifications on that body of work. The pilot study benchmarked against a high end cart system made by the same manufacturer, so the reference standard and the device under test share a vendor, which is a conflict worth naming even though the comparison direction disfavours the handheld. And rating studies of image quality are inherently subjective, which the authors of such work generally acknowledge.

For the models there is nothing. No validation of scene recognition, automatic spectrum tracing or automated bladder volume was located in any form, vendor or independent. The 2024 clearance summary states that clinical studies were not required to support substantial equivalence, which is a legitimate regulatory position for a device of this class and means no clinical performance record was generated for the automated features either.

Graded C: solid independent evidence that the instrument images well, none at all that the automation performs.

One secondary use pathway is disclosed and it is disclosed loosely. The manufacturer states that the phased array probe connects to hospital information systems and that de identified data is stored and shared to satisfy needs for teaching, training, consultation and case sharing. Naming a secondary use of exam data at all is better than the silence most device makers offer, and the stated purposes are educational rather than commercial.

Everything that would make the statement assessable is missing. No de identification standard is named, so a reader cannot tell whether images are stripped of embedded identifiers, whether the standard used is the safe harbour method or expert determination, or who verifies it. Where the data is stored is not stated. Who it is shared with, under what agreement, and whether a customer can decline are not addressed. Whether any of it informs development of the recognition models is neither claimed nor denied.

A second pathway is undescribed rather than loosely described. Activation of a transducer requires connecting to a network and entering a manufacturer issued code valid for 24 hours, so the device contacts vendor systems before first clinical use, and nothing published describes what is exchanged.

The architecture works in the buyer's favour by default. The application runs on hardware the customer owns and manages, exams live there, and the probe communicates with it over the customer's own wireless network, so a cautious organisation can keep the data path inside its own boundary. Graded C: a real disclosure with no definitions behind it, on an architecture that is otherwise conservative.

The manufacturer describes the right artefacts and publishes few of them, which is the difference between this record and the two large incumbents in the cluster.

What is stated is a commitment to transparency delivered through cybersecurity whitepapers, product user manuals, manufacturer disclosure statements for medical device security, software bills of material and assistance with deployment planning. A regional security page asserts compliance with the United States health privacy statute and the European data protection regulation. The disclosure statement is the document a hospital privacy office actually wants, because it answers protected data handling in a standardised comparable form.

The distinction that sets the grade is availability versus publication. Philips publishes a system and data security document for the specific device in this cluster and GE HealthCare operates a portal serving those statements. Here the categories are named as things the manufacturer provides, and no product specific statement or security document for this device was located publicly, so a buyer must request them. The one cybersecurity whitepaper found in full covers a different product line and carries a confidentiality marking.

The business associate position is unaddressed in anything located, and it is a live question because the vendor describes de identified exam data being stored and shared for teaching, training, consultation and case sharing, and because first activation of a transducer requires contacting the manufacturer's systems over a network. Neither pathway is documented in privacy terms.

Graded C: a described programme with the right components, no published product level artefact and no contractual position.

A described programme with the right components, missing the one thing that distinguishes the two large incumbents in this cluster.

What the manufacturer publishes is a security page committing to transparency through named artefacts: cybersecurity whitepapers, product user manuals, manufacturer disclosure statements for medical device security, software bills of material and deployment planning assistance. Behind that sit operational practices described in some detail, including a patch management strategy that monitors vulnerabilities in third party operating systems and assesses their impact on devices, proactive end of life letters when a product approaches the end of its serviceable life, and guidance on secure decommissioning and disposal. Quality systems are certified to ISO 9001 and ISO 13485 as declared in the regulatory record, which is quality rather than security assurance but does indicate an audited process environment.

End of life letters and decommissioning guidance deserve specific credit. They address the part of a device lifecycle that most manufacturers ignore and that leaves hospitals holding unsupported equipment carrying patient data.

What is missing is the public accountability layer. No coordinated vulnerability disclosure policy, no researcher reporting channel and no public security advisory archive were located. Both Philips and GE HealthCare publish years of advisories naming weaknesses in their own products and crediting the researchers who found them, and that history is the strongest available evidence that a programme operates rather than merely exists. No independent security certification such as an information security management standard was located either.

Graded B: real, documented practice, no public record of it being exercised.

A clean and conventional clearance record with no artificial intelligence specific authorisation, which is what separates this from the A grades in the cluster.

The device is cleared through the 510(k) route, with a 2024 submission building on an earlier clearance of the same product family as predicate. The summary describes a Track 3 device employing phased array probes, an application installable on iOS or Android, imaging modes including tissue harmonic imaging and tissue Doppler, and exam types spanning cardiac, superficial and peripheral vessel work, with acoustic output confirmed below the applicable limits. Quality systems are declared against the federal quality system regulation and against ISO 9001 and ISO 13485, which is more quality system detail than most summaries carry.

The submission states that clinical studies were not required to support substantial equivalence. That is a legitimate and common position for an imaging device cleared against a predicate, and it is recorded here because it is the reason no clinical performance data exists in the public regulatory record for this device or for its automated features.

Nothing located indicates a separate authorisation for scene recognition, automatic spectrum tracing or automated bladder volume, so those features appear to have reached the market as part of the device rather than through their own review. Competitors in this cluster hold feature specific clearances and one holds a De Novo authorisation that created a classification.

International presence is broad and the manufacturer sells worldwide, though geographic availability of this product line has shifted recently, with direct North American sales opening only in late 2025.

Nothing was located in any form. No model card, no statement of training data composition or provenance, no performance figures for any automated feature, no subgroup analysis by sex, age, body habitus or ethnicity, no description of how models are updated or revalidated, and no bias statement of any kind.

The absence is structural rather than an oversight in publishing. The automated features reached the market inside a device cleared on substantial equivalence to a predicate, with the clearance summary recording that clinical studies were not required, so no public performance record was generated for them at authorisation and none has been published since.

That matters more for two of these features than their modest framing suggests. Automated bladder volume is used to decide whether to catheterise, a decision with real consequences for infection risk, and volume estimation from recognised diameters is known to vary with body habitus and bladder shape. Automatic tracing of a Doppler spectrum produces resistance and pulsatility indices that a clinician may carry into a vascular assessment. Neither is a trivial number, and no evidence exists in public that either performs consistently across patient groups.

Graded D. This is the floor and it is the correct grade when nothing at all is disclosed, and it should be read alongside the low centrality grade: these are narrow models, and being narrow does not exempt them from the question.

Nothing published addresses what happens when an automated measurement is wrong. No performance warranty attaches to any automated output, no indemnity is described, and no adjudication or error reporting route specific to the automated features is offered.

The warranty position is the clearest in the cluster and it is entirely about hardware. Three years on the transducer covering normal wear and failure, one year on accessories including the charging cable and charging capsule, technical phone support within published hours, and an explicit exclusion of the customer's own mobile device. A buyer knows exactly what is covered if a probe fails and nothing about what is covered if a bladder volume reads low and a retention is missed.

Two features make the silence more pointed than the modest scope of the automation suggests. Bladder volume informs a catheterisation decision. Automatically traced Doppler indices feed a vascular assessment. Both are numbers a clinician may act on directly rather than impressions they will weigh, and no published position states who carries the consequence when either is wrong.

The conventional answer, that the clinician reading the image remains responsible, is more available here than on most records in this cluster, because every automated output appears on screen beside the anatomy that produced it and is inherently checkable. That answer is still not made anywhere located, and this axis records published positions rather than reasonable inferences.

Graded D on the same basis applied across this cluster.

Imaging connectivity is claimed at the usual level and one capability here is genuinely distinctive.

The manufacturer states integration with hospital information, radiology information and imaging archive systems, image and cine storage and export within the application, and de identified data sharing for teaching and consultation. No detail was located on how the integration is configured, what protocols it uses beyond the standard imaging format, or whether it requires additional software, so the claim sits at the level of assertion rather than specification.

The distinctive capability is within the vendor's own line rather than toward the record. The phased array probe docks into the manufacturer's trolley ultrasound system and operates as a transducer on it, so a department can buy one probe and use it at the bedside on a phone and in the room on a full system with the larger tool set. That is a real interoperability property, it removes a duplicate purchase, and no competitor in this cluster offers anything like it. It also creates a dependency: the fuller automated tools a buyer may want are reached through the vendor's own hardware rather than through open interfaces.

What is absent is the clinical record, in the same way as every other device in this cluster. No health level seven interface, no fast healthcare interoperability resources support and no named electronic health record integration were located. A bladder volume or a Doppler index reaches the archive and does not reach the note as a discrete coded result.

Deployment is simple and residency is undocumented, which is the cluster norm, with two specifics worth recording here.

The probe communicates wirelessly with an application on a phone or tablet the customer owns and manages under its own device policy, and exams live on that device. There is no vendor tablet to procure and no proprietary dock. Fast charging and full immersibility make the device practical in environments where a cabled or fragile alternative would not be, and the phased array probe can alternatively be driven from the manufacturer's trolley system, which changes the data path entirely for that mode of use and is not described in residency terms anywhere located.

The first specific is activation. A transducer cannot be used until it is activated with a manufacturer issued code over a wireless or cellular connection, and the code expires after 24 hours, so the device requires contact with vendor systems before first clinical use. Nothing published describes what that exchange contains or where those systems sit.

The second is manufacture and corporate location. The manufacturer is headquartered in Shenzhen, China, with North American operations in New Jersey, and the described pathway for de identified teaching and consultation data does not state which jurisdiction holds it. For an organisation with a data localisation obligation or a procurement policy that weighs country of manufacture, none of the information needed to make that assessment is published, and this index records that as an absence of disclosure rather than as a judgement about the manufacturer.

Graded C: an unusually clean local architecture, no published answer on anything that leaves the handset.

The clearest commercial proposition in this cluster, held below the top grade only because the figures themselves could not be verified from the vendor.

The structure is unambiguous and stated without hedging. Both probes are sold outright through the manufacturer's own online store, opened in North America in November 2025 specifically to make direct purchase possible. There is no subscription, and the manufacturer commits to that in plain language rather than leaving it to be inferred, describing the product as delivering everything up front with no additional fees. A standard three year warranty covers the transducer for normal wear and failure with one year on accessories, technical phone support hours are published, the exclusion of the customer's own mobile device from warranty is stated, and a demonstration programme lets a buyer trial before committing. Charging times, ingress rating and battery endurance are all published. That set of terms answers the total cost of ownership question more completely than any competitor here except Clarius.

What could not be established is the price. Vendor store pages did not yield retrievable figures, and the only numbers located come from a third party analysis quoting roughly $4,125 for the whole body probe and roughly $5,422 for the phased array, both explicitly converted from euro listings at a time when the product was described as available only in Europe and the United Kingdom. Under this index's currency convention a converted figure is not recorded as a dollar price, so the pricing record carries no numeric entry price.

One gap qualifies the headline. Some smart tools are described as optional one time upgrades, and no price for any of them was located, so the no additional fees commitment covers the base device rather than every capability a buyer might want.

Two probes cover the range that competitors need two or three to reach, and the environmental specification extends the settings further than most.

The whole body probe combines convex and linear modes in one housing, carries more than 20 presets and 22 exam types across abdomen, small organ, vascular, obstetrics, gynaecology, musculoskeletal, cardiac and FAST, and scans to 40 cm, which is deeper than several competitors reach. The phased array probe uses software based beam steering to simulate a convex footprint and is aimed at emergency and critical care cardiac assessment, covering parasternal, apical and subcostal views for pericardial effusion, left ventricular function and chamber size.

Environmental capability is where this pulls ahead. Both probes carry an IP68 rating and are fully immersible in compatible disinfectants, which admits them to operating rooms and intensive care units where high level disinfection is required and where a device that cannot be immersed is simply not usable. Battery endurance of roughly 120 minutes with a 35 minute fast charge supports continuous shift use, and the named settings run across acute care, primary care, private practice, education and remote medicine.

The phased array probe also docks into the manufacturer's trolley system, so the same purchase serves a bedside handheld role and a transducer role on a full system. No other record in this cluster offers that.

The artificial intelligence coverage is far narrower than the device coverage, reaching abdomen, thyroid, carotid and bladder only, and that limitation is recorded on the centrality and transparency axes rather than here, since the instrument does cover the settings it claims.