Infervision

Deep learning image analysis is the whole company across an unusually wide product line: InferRead covering lung CT, chest radiography, coronary CTA, and target reconstruction, InferOperate covering surgical planning for thorax, liver, and urology, and InferCare RECIST for standardized tumour assessment. No scanner or hardware of its own. The breadth is algorithmic breadth rather than a platform business wrapped around models.

Cleared to support CONCURRENT reading, meaning output is available alongside the radiologist's initial interpretation rather than after it. That is the same design choice as Riverain and the opposite of Ibex's second-read sequencing, and it carries the same tradeoff: faster workflow, greater exposure to automation bias.

An independent reviewer characterised the tool as a lung nodule spell-checker, which is a fair description of the intended relationship, the human reads and the algorithm catches misses. The surgical planning products sit further from autonomy still, producing plans a surgeon evaluates.

Capability is described concretely, including automated lung segmentation, nodule identification and labelling by type, detection down to 4 mm, and automated lung density analysis, and the regulatory scope of each clearance is stated. What is not published is performance data: no sensitivity, specificity, reader study results, or per-nodule-type accuracy were located for any product in the line. For a vendor claiming detection at 4 mm, where false positive rates rise sharply, the absence of published performance is the material gap.

Nothing was retrieved on retention of imaging or derived outputs, de identification, training use, or deletion at contract end, and no model, hosting arrangement or sub processor list was named for any market. The training question is unusually pointed because of how the portfolio has grown, spanning nodule detection and characterisation, chest radiography, coronary imaging, reconstruction and a surgical planning line across three organ systems.

Building and extending that range required substantial and varied imaging data, and whether any came from customer deployments, and whether a health system can decline to contribute, is unstated. The screening programmes deserve separate consideration rather than being folded into the general case.

Deployments procured through international agencies for tuberculosis screening operate in settings where participants may have limited practical ability to refuse a scan, where local data protection enforcement may be weaker, and where the programme rather than the individual is the decision maker.

Data governance obligations do not lessen in those settings, and a vendor's published position becomes more important rather than less when the usual counterweights are absent, because there is no procurement office reading the terms and no regulator likely to ask. Published material addresses those deployments no more than the commercial ones. Ask what is retained after a read in each market, whether imaging trains models, and what governs data in agency procured screening.

Deployment scale is substantial and long running, reported at more than 380 hospitals and imaging centres globally processing over 55,000 cases daily, with five years of international clinical use cited at the time of first FDA clearance and a named US partnership with SimonMed, one of the largest outpatient imaging providers.

Real world tuberculosis screening deployments in Kyrgyzstan and Zimbabwe, including UN procurement for Eastern Europe and Central Asia, demonstrate the product functions in low resource settings. What is missing is peer reviewed validation: no published reader study, diagnostic accuracy paper, or outcome evidence specific to these products was located, which is a notable gap given the deployment volume. Scale of use is not evidence of benefit.

Converted from Not Rated. No stewardship terms were located, and the deployment breadth makes this a per market question with no published answer for any market.

Nothing was retrieved on retention of imaging or derived outputs, de identification, model training use, or deletion at contract end.

The training question is unusually pointed because of how the product portfolio has grown. The suite now spans lung nodule detection and characterisation, chest radiography, coronary imaging, target reconstruction and a surgical planning line across three organ systems. Building and extending that range required substantial and varied imaging data. Whether any of it came from customer deployments, and whether a health system can decline to contribute, is unstated.

The screening programmes deserve separate consideration rather than being folded into the general case. Deployments procured through international agencies for tuberculosis screening operate in settings where participants may have limited practical ability to refuse a scan, where local data protection enforcement may be weaker, and where the programme rather than the individual is the decision maker. Data governance obligations do not lessen in those settings, and the published material addresses them no more than it addresses the commercial ones.

The regulatory footprint is genuinely broad, spanning five jurisdictions, and none of those authorisations speaks to data handling.

Ask what is retained after a read in each market, whether imaging trains models, and what governs data in agency procured screening deployments.

Converted from Not Rated. No published position was located despite United States operations and a named imaging partnership there.

No business associate agreement, addendum, role statement, subcontractor flow down, breach notification timetable or review cadence was retrieved.

The role is not ambiguous. A United States hospital or imaging provider is the covered entity, and a vendor receiving studies and returning findings processes protected health information on its behalf as a business associate with direct liability.

The structure is what a buyer must pin down. The company is China headquartered with a separate United States presence and clearances in five jurisdictions. Which legal entity signs, whether the United States entity is the contracting party and is capitalised and staffed to carry the obligation, whether any processing or support access occurs outside the country of care, and how affiliate access is controlled are all questions a single agreement has to answer. Where engineering or support touches live studies from another jurisdiction, the subcontractor terms are the only control.

This index asks the same of every non domestic vendor it holds, and several have the same gap. What differs here is that two regulatory frameworks with express cross border provisions apply simultaneously, so the agreement has to be coherent under both.

Ask which entity signs, whether processing and support remain in country, and for the affiliate and subcontractor terms in writing.

Converted from Not Rated. No independent security attestation was located across two differently phrased searches.

No SOC 2 of either type, no ISO 27001, no HITRUST, no trust centre and no penetration testing statement was retrieved.

What is published instead is the broadest multi jurisdiction regulatory footprint of any vendor in this lane: authorisations from the United States, European, United Kingdom, Japanese and Chinese regulators. Assembling five regulatory dossiers is substantial work and it demonstrates a functioning quality system. It is not an information security attestation, and this index has now drawn that line consistently across imaging vendors from six countries. Device authorisation examines whether the software is safe and performs as labelled. It does not examine how patient studies are protected.

The absence carries more weight here than for a single market vendor. A buyer's security review normally leans on a recognised third party report precisely because it cannot inspect the vendor directly, and that reliance matters most where the vendor is furthest away, operating under a different legal system, and where enforcement of a contractual promise would be practically difficult. An attestation is a partial substitute for reach.

Multiple peers in this lane hold ISO 27001 and are credited for it, including vendors of comparable size and international spread, so this is achievable rather than unusual.

Ask what independent security assessment exists, under which standard, and whether the United States entity is separately in scope.

The broadest multi jurisdiction footprint of any vendor in this index's imaging lane, spanning US FDA, European CE, UK UKCA, Japan PMDA, and China NMPA. US clearances include InferRead Lung CT.AI first cleared in 2020, a further clearance in May 2025 for enhanced features including 4 mm nodule detection and automated lung density analysis, and a separate clearance for InferCare RECIST for standardized tumour assessment.

European and UK marking covers InferRead DR Chest under MDR and UKCA, with multiple Class III NMPA registrations in China. Clearing five distinct regulatory regimes is materially harder than clearing one, and the iterative US clearances show ongoing maintenance rather than a legacy authorization.

Converted from Not Rated. The prior analysis named the right question and it is one of the better governance observations in this index, so it is worth stating at full strength.

No governance framework, monitoring commitment, subgroup performance analysis or bias evaluation was located.

The concern is a population mismatch between where the models were built and where they are used. Development drew substantially on Chinese patient populations. Deployment spans United States and European health systems and, distinctly, tuberculosis screening programmes in Central Asia and elsewhere procured through international agencies. Those populations differ in disease prevalence, in comorbidity profile, in body habitus, in imaging equipment and in acquisition protocol.

Prevalence is the part that matters most and is least intuitive. A detection model's positive predictive value depends on how common the finding is in the population being scanned, so a model validated where tuberculosis prevalence is one thing behaves differently where it is another, even with identical sensitivity and specificity. In a screening programme, where most people scanned are well, that shift determines how many healthy people are pulled into follow up.

The screening deployments are also the ones with the least capacity to detect a problem locally, since they run in settings chosen precisely because radiologist availability is limited.

Ask for performance stratified by deployment region and by scanner type, and what post deployment monitoring exists in screening programmes.

Capability is described concretely, covering automated lung segmentation, nodule identification and labelling by type, detection down to four millimetres and automated density analysis, with the regulatory scope of each clearance stated. What is not published is any performance data at all: no sensitivity, specificity, reader study results or per nodule type accuracy were located for any product in the line, and no warranty, indemnity or remediation commitment attaches.

The detection threshold is what makes that absence material rather than routine. False positive rates rise sharply as nodule size falls, and a four millimetre threshold sits well inside the range where benign findings vastly outnumber malignant ones, so a product claiming detection at that size is claiming the hardest part of the problem and publishing nothing about its cost.

The clinical consequence is concrete: a flagged sub centimetre nodule initiates surveillance imaging, and in most cases that surveillance ends in nothing while the patient carries months of uncertainty and additional radiation. A department needs the positive predictive value at the stated threshold in a population like its own before it can judge whether the sensitivity is worth having, and neither number exists. Ask for sensitivity and positive predictive value by nodule size band, the false positive rate per study, and what proportion of flagged findings resolved as benign.

Integration into the image review workflow is emphasised as a core design priority alongside speed and accuracy, and the product is described as workflow friendly and integrating with existing review processes. Deployment across more than 380 sites in multiple countries, plus a partnership with a large US outpatient imaging provider and UN procured deployments, indicates the integration generalises across varied PACS environments rather than requiring bespoke work. No published connector list or API documentation was located. Radiology workflow integration; no EHR integration claimed.

Converted from Not Rated. No hosting, region, tenancy or residency terms were located, and the cross border question here has two sides rather than one.

Nothing published states the cloud provider, the region, whether an on premise option exists, how tenancy is separated, or what the subprocessor chain is. Several peers in this lane do publish an on premise option and are credited for it, so the disclosure is achievable in the category.

The jurisdictional analysis needs stating carefully and evenly, since this index applies the same treatment to vendors headquartered in France, Korea, Israel, Australia, Canada, India and Lithuania. Any vendor processing patient data outside the country of care raises the same three questions: which legal entity contracts, where processing occurs, and whether support or engineering personnel access live data from another jurisdiction.

What is specific here is that the flow runs in both directions under distinct regimes. Patient imaging from United States and European health systems moving toward a China headquartered organisation engages those buyers' own transfer rules. Separately, China's data security and personal information protection laws impose their own requirements on data held by entities there, including provisions on government access and on cross border transfer out. A buyer's diligence has to satisfy both frameworks, not one.

An on premise deployment, if available, would resolve most of it.

Ask which entity contracts, where inference runs, whether local deployment is offered, and what access exists from outside the country of care.

No rates are published, though the company has stated it offers the system under a number of pricing options rather than a single model, which at least signals flexibility. UN procurement for tuberculosis screening deployments implies the product can be bought through public tender processes, which typically involve competitive pricing disclosure to the purchaser even where rates are not public. Nothing on per study, per site, or licensing cost was located.

The widest coverage in this index's imaging lane, spanning three distinct clinical functions rather than one. Diagnostic detection covers lung CT, chest radiography, and coronary CTA. Surgical planning covers thorax, liver, and urology, a domain almost no imaging AI competitor addresses. Oncology response assessment is covered by InferCare RECIST.

Settings span diagnostic radiology, lung cancer screening programmes, surgical planning, oncology follow up, and public health tuberculosis screening in low resource countries. That last dimension, functioning in UN procured screening programmes in Kyrgyzstan and Zimbabwe, extends relevance well beyond the western hospital market most indexed competitors serve.