Ultromics

The model is the product and it does something a person cannot. A cardiologist reading an echocardiogram is not attempting to determine whether amyloid protein is depositing in the myocardium, because that signal is not available to human reading of the image; the usual route to that answer involves scintigraphy or biopsy.

The company frames its own founding around this distinction, describing a deliberate move beyond automating measurements toward identifying disease directly from routine imaging. That is the same reasoning this index used to grade the electrocardiogram algorithms at A, and it applies here for the same reason: an algorithm that extracts a signal invisible to the reader is a stronger centrality case than one that speeds up a task a reader already performs.

A screening tool with its boundaries drawn tightly and stated openly. The output identifies patients who may have the disease and could benefit from further diagnostic testing; it does not diagnose, and the confirmatory pathway remains what the guidelines specify.

The indication is unusually specific for this index: adults aged 65 and over with heart failure undergoing echocardiographic assessment. Naming the population rather than the modality is a real constraint on where the model may be applied, and it means a buyer can tell when they are outside the tested envelope.

Held at B because nothing published describes what a clinician sees when the model is uncertain, or how a negative result should be weighed against clinical suspicion, which matters for a screening test whose miss is a delayed diagnosis of a progressive disease.

The most complete performance disclosure encountered in this sweep, and one specific choice earns the grade.

The company publishes two different sets of numbers. The study figures from the peer reviewed validation are 85 percent sensitivity and 93 percent specificity. The figures from clinical evaluation under intended use conditions, taken from the clearance summary, are 84.5 percent sensitivity and 89.7 percent specificity. Publishing the lower, real conditions number alongside the headline is rare: most vendors quote the flattering figure and stop. Pointing the reader to the clearance number for full results goes further still, because that document is public and not written by the company.

Around it sit the cohort size, the disease prevalence in that cohort, the area under the curve, the exact input required and the site count. What is not published is the architecture or the training data composition.

Nothing identifies any party in the chain: no model or model family, no hosting provider, no sub processor list was located in two passes, and no retention schedule, encryption detail or position on model improvement was found. The architecture makes that absence consequential rather than routine.

The platform is cloud based and processes echocardiographic studies from customer sites, so images leave the hospital by design, which is a materially different exposure from the on device inference credited elsewhere in this index and it means every unnamed party in the path handles moving pictures of a patient's heart alongside the identifiers attached to them. The corpus raises the second question.

The company describes building on one of the largest echocardiography datasets in the world, and a dataset of that scale came from somewhere: whether it was assembled under research arrangements separate from commercial deployment, or whether customer studies contribute, is unstated in either direction, as is whether a site can decline while continuing to use the product.

This index has recorded that the two are usually separate and that the distinction is precisely what a buyer needs stated rather than assumed. Ask which cloud hosts it and in which region, for a sub processor list, what is retained after a read, and whether processed studies feed the dataset.

Deep, external and now embedded in guidance, which is this index's top bar.

The amyloidosis validation was multi centre and international, run with Mayo Clinic and investigators at another academic centre across more than 18 sites, published in a leading cardiology journal, with an external cohort of 2,719 patients and more than 12,500 cases in total. Performance held across every major subtype of the disease and, importantly, discriminated it from the conditions most often confused with it.

The heart failure product clears the guideline bar directly: evidence supporting it is cited in the 2025 diastology guidelines of the American Society of Echocardiography. More than 25 peer reviewed studies support the platform overall.

One honest qualification. Guideline citation of supporting evidence is not the same as a guideline recommending the test by name for a specific decision, which is the highest tier this index recognises. This sits just below that and well above vendor publication.

Graded on an honest basis. No retention schedule, encryption detail or model improvement position was located in this pass.

The platform is cloud based and processes echocardiographic studies from customer sites, so images leave the hospital, which is a different exposure from the on premise inference used by the ambient monitoring vendors in this index. The company also describes building on one of the largest echocardiography datasets in the world, which raises the usual question of what contributed to it and on what terms.

Graded on an honest basis and flagged for re verification. No compliance statement or agreement posture was located in this pass.

The company is British with a United States commercial footprint and states availability in the United States only, so the arrangements exist and were not retrieved. A buyer should establish where processing occurs given the cloud architecture and the Oxford base.

Recorded honestly and provisionally: the dedicated trust and security search this index requires was not run in this pass, and no attestation was encountered incidentally.

As with every cleared software device in this index, quality system obligations subject to inspection do apply, and they cover design controls and manufacturing quality rather than information security. Do not read regulatory rigour as a security posture.

Among the strongest regulatory positions in this index, and notable for how it was obtained as well as what it covers.

Four clearances and two Breakthrough Device designations, with the amyloidosis product the first artificial intelligence tool cleared for that condition and cleared under a published number a reader can look up. It was also the first device enrolled in the agency's Total Product Lifecycle Advisory Programme to reach marketing authorisation, out of a pilot cohort of fifteen breakthrough cardiovascular devices, which makes it a reference case for that pathway rather than merely a participant in it.

The designations converted, which this index treats as the question that matters: holding a Breakthrough designation is not clearance, and here two designations sit alongside four actual clearances.

One pairing worth noting for comparison. Another indexed vendor holds the first clearance for detecting the same disease from a standard electrocardiogram, and was also in the same advisory programme pilot. Two cleared artificial intelligence screens for cardiac amyloidosis now exist on two different routine tests.

Better than most in this index and short of complete.

What is disclosed is substantive. The external validation is described as multi ethnic and international across more than 18 sites, and performance is reported as consistent across all major subtypes of the disease. Most usefully, the model was tested against the conditions that mimic amyloidosis on an echocardiogram, hypertensive heart disease, heart failure with preserved ejection fraction and hypertrophic cardiomyopathy, which is the discrimination that determines whether a screening tool creates false alarms in exactly the patients who present similarly.

What is missing is the numbers behind the diversity claim. Multi ethnic is asserted without a breakdown, and no performance is reported by ethnicity, sex or site. That matters here because amyloidosis subtypes are not evenly distributed: the hereditary transthyretin variant most common in the United States is carried disproportionately by people of West African descent, so a model whose sensitivity varied by ancestry would miss the disease in the group where one major subtype concentrates.

This company publishes the number that makes it look worse, and that single choice earns the grade. Two sets of figures appear. The peer reviewed validation reports eighty five per cent sensitivity and ninety three per cent specificity. The clinical evaluation under intended use conditions, taken from the clearance summary, reports eighty four point five and eighty nine point seven.

Publishing the lower real conditions figures alongside the headline is rare, because most vendors quote the flattering number and stop, and the gap between a controlled study and intended use is exactly where a buyer's own experience will land. Pointing a reader to the clearance summary for full results goes further still, since that document is public and was not written by the company, so the vendor is directing diligence toward a source it does not control.

Around those figures sit the cohort size, the disease prevalence in that cohort, the area under the curve, the exact input required and the site count. Prevalence deserves particular note: without a base rate a sensitivity figure cannot be converted into the probability a clinician actually needs, and almost nothing else in this index supplies it.

Held below the top grade because architecture and training data composition are unpublished and no warranty, indemnity or remediation commitment attaches. Ask for performance in a population with your own prevalence, and what happens on inadequate image quality.

Designed to disappear into the existing workflow rather than sit beside it. Studies are analysed automatically and results return into existing clinical systems with no additional scanning steps and no change to how the echocardiogram is acquired, typically within 20 minutes so the finding is available during routine interpretation rather than after it.

That timing is the operative design point: a result that arrives after the report is signed changes nothing. The workflow extends through to automated billing, which closes the loop on the reimbursement pathway. Held at B because no interface standard, image archive certification or named record system integration was located.

Cloud based and explicitly designed to scale from a single echocardiography laboratory to an enterprise, which suits a screening product whose value rises with the proportion of studies it sees.

Held at B because no hosting region, retention schedule or customer controlled option is published, and because the company states availability in the United States only while being based in the United Kingdom, so where studies are processed is a question a buyer will ask and public material does not answer.

No list price, and the payment mechanism is published with the actual codes, which is the pattern this index has found repeatedly among reimbursed diagnostics and consistently credits.

The heart failure product carries a Category III procedure code for outpatient use and a new technology add on payment for inpatient settings, both named. That lets a hospital work out what it will be paid before it works out what it will pay, which is the more useful half of the calculation. The workflow includes automated billing, so the reimbursement loop is closed inside the product.

What is absent is the vendor's own charge and whether the amyloidosis product has an equivalent pathway, which a buyer should establish separately since the two products were cleared years apart.

Deliberately narrow. One modality, echocardiography, and two conditions, heart failure with preserved ejection fraction and cardiac amyloidosis, with the amyloidosis indication further restricted to adults aged 65 and over presenting with heart failure.

That focus is the source of the depth elsewhere on this record and it bounds the reach. Deployment is in echocardiography laboratories and cardiology services at named academic centres, and the company states availability in the United States only despite being based in the United Kingdom, which is an unusual restriction and limits the setting further.