Clarius Mobile Health

A portfolio of cleared models on a hardware business that works without them. Clarius sells seven wireless handheld scanners, and the company states plainly that the scanner works for core imaging without a membership. Clarius Intelligence adds measurement and guidance models, each cleared on its own: MSK AI, Bladder AI, OB AI for fetal biometry, Prostate AI, Median Nerve AI, Ejection Fraction AI and automatic preset selection, plus T-Mode anatomy visualization for education and procedural planning.

These are real measurement models producing numbers a clinician acts on, and several are included with membership rather than sold as the product. Remove every model and a differentiated ultrasound business remains, so the AI here is a feature layer on a product whose value stands without it.

The models measure and the clinician decides, with visible calipers and no published behavior at the edges. When a model is activated during scanning it highlights anatomy, places calipers and shows the measurements on screen before the clinician adds them to the record; Ejection Fraction AI calculates ejection fraction from four standard views hands-free, Bladder AI calculates volume, and MSK AI guides the operator to standardized views.

The clinician can see where each caliper landed, which is real oversight for a measurement tool. The users are often new to ultrasound: OB AI is aimed at midwives, nurses and new users in resource limited settings, where an operator may not recognize a plausible but wrong measurement. Nothing on the site says what a model does with an image that lacks the expected anatomy, whether it declines to measure, or what confidence signal the operator sees.

Each model is named by task and scanner; how the models work and how well is not described. Clarius lists its AI models with the scanners they run on and the specialties they serve (Prostate AI on the C3 and EC7, Median Nerve AI on the linear scanners, Bladder AI and Ejection Fraction AI on the curved and phased arrays, OB AI on the C3, MSK AI on the linear scanners, Auto Preset AI), describes them as deep learning, and gives the OB AI training corpus as more than 30,000 de-identified images.

Device specifications and encryption parameters are published in detail. There is nothing on architecture, training data beyond that one count, accuracy or agreement figures, or model versions, and nothing on how models delivered through the app are updated and recorded against the measurements they produced.

A short chain for the models, with the cloud host unnamed. Each AI model is Clarius's own cleared device (six AI clearances in its name from 2023 to 2026), runs in the Clarius app with its scanners, and changes only through the company's releases, so no outside model provider sits between an image and a measurement. Studies can stay on the device; cloud data rests in Canada or a region the customer selects.

The privacy policy names web analytics and app platform providers for the website and apps, but not the cloud infrastructure provider for patient data or any subprocessor list. An in house, cleared model chain with the hosting layer unnamed.

Scale and one independent hardware comparison are published; no accuracy figure for any model is. Clarius reports more than 7 million scans in over 70 countries and says its AI is backed by peer reviewed studies and real world use. The independent study it cites (Park and colleagues, Ultrasound, 2023) compared image quality across five point of care devices and rated its high frequency scanners highest; that is evidence about the hardware, not the models.

For OB AI the company cites a midwife sonographer's testimony on caliper placement while teaching. No sensitivity, agreement with manual measurement or outcome figure is published for the fetal biometry, bladder, prostate, median nerve, ejection fraction or MSK models, and no named institution reports measured results. The clearances required performance data, which sits in the FDA summaries rather than on the company's pages. Scale and testimony, no method or denominator.

Storage choices and the separation of identifiers are stated; training use of customer images is not. The privacy policy says examination information (images, cine loops, measurements) is stored separately from user and patient identifiers, that this de-identified examination information may be transmitted unencrypted, that cloud data sits in Canada or in a region the customer selects, and that customers can keep studies on the device instead.

OB AI is stated to have been developed on more than 30,000 de-identified fetal images. The policy says the company uses machine learning to analyze sales and support communications, but does not say whether customer examination images are used to train or evaluate models, whether that differs between local and cloud storage, or whether a customer can decline; nor does it give a retention period for examination data. Safety engineering, meaning how a wrong measurement is handled as a safety event, is not published. Specific commitments on storage and de identification, with training use and retention missing.

The privacy policy carries the business associate commitment, and the agreement itself is not published. Clarius's privacy policy (updated May 2026) covers the product, the app and Clarius Cloud, and says patient information is used and disclosed under the terms of acquisition, the policy, the law, or additional written agreements, naming a business associate agreement for US users and a data protection addendum for EU users.

It describes how examination information is stored separately from identifying information, and where cloud data rests: in Canada for institutions created before app version 4.2.1, and in a region the customer selects for later institutions. The Cloud page describes HIPAA compliant patient and exam management with encrypted storage, and studies can also stay on the device. The privacy document is substantive, though neither the agreement nor its tier is published. Scanners are also sold to individual clinicians through the online store, and nothing says whether a sole practitioner receives the business associate agreement or what it covers.

Controls are described in detail with nothing independent behind them. Clarius publishes the encryption used on the scanner's wireless link and for device pairing, a feature that removes protected health information, and on its Cloud page describes HIPAA compliant, encrypted storage. That is more technical detail than a compliance badge, and it lets a reviewer judge the choices.

What is missing is any outside check: no SOC 2 report, ISO 27001 or other certification, independent penetration test or trust portal is named on the AI, Cloud, privacy or reference pages, and the company's device clearances cover the scanners and models rather than its information security. Controls asserted with no independent check. Scanners are also bought by individual clinicians who cannot run a vendor security review.

Each AI model carries its own FDA clearance, and the list can be checked. FDA records show Clarius's AI clearances as K222406 (Clarius AI, the MSK model, January 2023), K232257 (Bladder AI, November 2023), K233955 (OB AI, June 2024), K243853 (Prostate AI, April 2025), K250226 (Median Nerve AI, May 2025) and K253593 (Ejection Fraction AI, March 2026), alongside the scanner clearances (most recently K232704, October 2023).

The company is Health Canada licensed and CE marked, says regional clearances vary by AI model, publishes a market clearances list on its support site, and marks models such as Prostate AI as available in approved regions only. Its LLM reference page summarizes the position as FDA 510(k) cleared, which is true per model rather than for every feature: T-Mode anatomy visualization is described for education and procedural planning.

Nothing is published on how model behavior is governed or tested across patients. OB AI is described as developed on more than 30,000 de-identified fetal ultrasound images, with no statement of where they came from, which growth reference converts measurements into age and weight, or how performance varies by gestational age, maternal body habitus or population.

That matters most for this model: it is aimed at midwives and nurses in resource limited settings, the populations most likely to differ from a development set, and a dating error changes decisions about delivery timing and growth restriction. The bladder, prostate, median nerve and ejection fraction models raise the same question for body habitus and sex specific anatomy.

No governance framework, subgroup result or model documentation appears on the company's AI, privacy, Cloud or reference pages; the FDA summary for each model (for OB AI, K233955) is where subgroup results would appear if reported. D.

Mechanisms let the operator check a measurement; nothing stands behind the output. Calipers placed by a model appear on the image before the clinician adds the measurement to the record, so a misplaced caliper is visible and can be rejected. The company publishes hardware terms (a Clarius Care warranty program and an online store), but nothing attaches to a model's measurement: no accuracy commitment, indemnity or remediation.

No error rate for any model appears on the company's pages; the FDA summary for OB AI (K233955), the most prominent model, is where one would appear. For a fetal biometry tool used by new operators, the published material is silent on what the contract provides for a wrong estimate and on what agreement with expert measurement the clearance showed.

Getting measurements into the record is claimed; the systems and standards are not named on the company's own pages. Clarius says measurements from its AI models can be added to the patient record from the app, and Clarius Cloud lets users manage exams, build and export PDF reports, and share studies with colleagues, with studies stored on the device or in the cloud.

The site names no record system, PACS vendor, interface or DICOM conformance statement, and nothing says whether results flow back from an order or whether the connection is two way. The membership pages may describe DICOM and PACS export in more detail.

The customer chooses where studies rest, and cloud residency is stated. Studies can stay on the acquisition device or go to Clarius Cloud. The privacy policy says cloud data for institutions created before app version 4.2.1 sits in data centers in Canada, and for later institutions in a geographic region the customer selects when creating the institution, while some usage and account information is held separately.

The scanners connect directly to a phone or tablet over their own wireless link, so imaging does not depend on the facility network, which matters in the low connectivity settings the product targets. The company names no cloud infrastructure provider and lists neither the regions offered nor any tenant isolation, recovery or availability commitments.

Prices are published on the company's own store, with a route to buy without a sales conversation. Clarius's online store lists each scanner with its price, from $4,490 (for example the curved, phased array and endocavity models) to $6,090 (the ultra high frequency L20 and dual array models), each including one year of the Clarius Essentials membership, and sells accessories online; separate regional stores serve Canada, the EU, the UK and Australia.

The company states that the scanner works for core imaging without a membership, so the recurring fee buys cloud storage, AI models and team tools rather than the right to use hardware already paid for, and it publishes a Clarius Care warranty program. The store does not show the membership renewal price after the first year, and enterprise or volume pricing runs through a quote. Published prices, a stated unit of charge (per scanner with a year of membership) and a route to start without contacting sales.

Coverage is built scanner by scanner and model by model, with validation behind the cleared models. Seven scanners cover curved array abdominal, obstetric and cardiac work, phased array cardiac and eFAST, linear vascular, musculoskeletal and nerve imaging, ultra high frequency aesthetics and superficial structures, and endocavity gynecology, IVF and urology; there is a veterinary line as well.

Buyers include emergency and critical care, primary care, obstetrics and midwifery, musculoskeletal and sports medicine, urology, aesthetics and plastic surgery, medical education and global health, in more than 70 countries. Each AI model is cleared for its own task (bladder volume, fetal biometry on the C3 scanner, prostate, median nerve, ejection fraction, MSK views), and the company says regional clearances vary by model.

Validation behind each model sits in its clearance; no published study shows where the handheld and its models stand against cart based systems by indication. Named coverage with validation behind part of it.