GuideAI Health

Detection is the product and it operates on scans that already exist. The software applies AI models to lower extremity CT to identify suspected vascular occlusion and flag it for prioritized radiologist review, with the company's stated premise that early vascular changes are subtle, distributed, and easy to overlook even with modern imaging. There is no non AI version of finding a subtle stenosis pattern across a routine study.

The regulatory classification defines the oversight model precisely, which is why this grades well despite the company being early. It is cleared as a computer aided triage and notification device, meaning it reorders the radiologist's worklist rather than rendering a finding: it flags suspected cases for prioritized review inside existing imaging workflows. The radiologist still reads and diagnoses every study.

Triage and notification is the most conservative category of imaging AI clearance, and choosing it rather than a detection claim is a deliberate scoping decision a buyer should understand as a feature.

Performance is published with the specificity the clearance required: 95 percent patient level sensitivity in 2D analysis and 94 percent in 3D analysis for identifying peripheral artery disease positive patients, with the positive definition stated explicitly as at least one lesion at 50 percent or greater stenosis. Publishing the case definition alongside the sensitivity figure is what makes the number interpretable. Specificity, model architecture, and validation cohort composition are not published.

No published handling or governance disclosure was located, no model or hosting arrangement named, and no terms describing where inference runs or what is retained were found. Two specifics make the questions concrete rather than generic.

A triage device analysing lower extremity computed tomography operates on the full study rather than on a derived summary, and those are large volumetric datasets, so whatever the architecture turns out to be, complete diagnostic imaging is moving somewhere on every case rather than a measurement or a thumbnail.

Establish whether any of it is retained after the triage decision is returned and for how long, because the department needs the result rather than the copy, and a vendor holding full studies has an asset the customer never intended to transfer. Second, the clearance rested on clinical performance testing across a patient dataset, and nothing states where that imaging came from, under what consent or authorisation, or whether the company retains it.

For a company founded by radiologists the likeliest provenance is institutional data, which carries its own conditions on secondary use, and a buyer being asked to contribute studies to future development should establish what basis the existing corpus was assembled under before agreeing to add to it. Ask for retention after triage, the corpus provenance, and whether customer studies inform development.

Evidence today is the clinical performance testing submitted for clearance, which is real but narrow: sensitivity figures on identifying PAD positive patients, with no published prospective deployment data, reader study, or outcome evidence that earlier flagging changes time to intervention.

The company's stated rationale, that peripheral vascular disease is too often missed or detected late with consequences up to limb loss, is clinically sound and widely accepted, but the link from triage flag to averted amputation is asserted rather than demonstrated. The company is very early, having received clearance in mid 2026.

No published handling or data governance disclosure was located, and no terms describing where inference runs or what is retained were found.

Two specifics make the questions concrete rather than generic. A triage device analysing lower extremity CT operates on the full study rather than a derived summary, and those are large volumetric datasets, so whatever the architecture turns out to be, complete diagnostic imaging is moving somewhere on every case. Establish whether any of it is retained after the triage decision is returned, and for how long, since the department needs the result rather than the copy.

Second, the clearance rested on clinical performance testing across a patient dataset, and nothing states where that imaging came from, under what consent or authorisation, or whether the company retains it. For a company founded by radiologists, the likeliest provenance is institutional data, which carries its own conditions on secondary use. A buyer contributing studies to any future development work should ask what basis the existing corpus was assembled under before agreeing to add to it.

No commitment was located in public materials. Two structural facts make this worth establishing early rather than assuming it will be handled at contracting.

The first is the device's position. A triage and notification product has to receive the imaging study to analyse it, and it operates inside the hospital's radiology workflow, so the vendor is handling protected health information from the first installation. An agreement is required in practice and none is described, priced or offered publicly.

The second is corporate structure, which changed recently. The operating business is in Boston, but the listed entity is a British Columbia corporation with a registered office in Vancouver, taken public through a plan of arrangement in June 2026. That places one company under two privacy regimes at once: the United States framework governing the clinical deployments, and Canadian federal and provincial privacy law reaching the corporate parent. Nothing published addresses either, or how patient data moves between the operating and parent entities.

Ask whether the agreement is executed by the United States operating entity or the parent, and whether any patient data reaches the parent for any purpose.

No independent attestation was located, and no security page or trust centre exists. FDA clearance addresses device safety and effectiveness and is not a substitute for an information security attestation, which is the distinction this row most often has to make.

There is, however, a specific place a buyer should look that the company's website does not advertise. This is now a reporting issuer: GuideAI Health Corp. is a British Columbia corporation listed on Cboe Canada, and it filed a final non offering prospectus dated 12 June 2026 with the securities regulators of British Columbia, Alberta and Ontario in connection with the business combination that took it public. A prospectus carries a risk factor section, and for a company whose product processes patient imaging that section has to address data security, privacy and regulatory exposure in terms the company is legally accountable for.

So the disclosure a security reviewer wants may exist in a filing rather than on a marketing page. It is on the company's SEDAR+ profile, and the company states a copy can be obtained without charge by email. Graded C because nothing is published on any surface a buyer would ordinarily check and the filing was not retrieved here, but a diligence process should read the prospectus risk factors before concluding the company has said nothing.

FDA 510(k) cleared as a Class II Software as a Medical Device, specifically as a computer aided triage and notification device for flagging suspected vascular occlusion on lower extremity CT, with supporting clinical performance data of 95 percent patient level sensitivity in 2D and 94 percent in 3D. For a company at this stage, holding a cleared product with a published performance basis and a precisely scoped indication is a strong position. Buyers should read the indication narrowly: it is triage of lower extremity CT for suspected occlusion, not general vascular assessment, and the company's broader vascular ambitions are not cleared.

No governance framework or bias evaluation was located. Peripheral artery disease prevalence and presentation differ meaningfully by diabetes status, race and smoking history, and CT protocols vary by site, so subgroup performance is a live question that published materials do not address.

The sharper point is what the company chose to publish about its own performance. Clearance materials are quoted as reporting 95 percent patient level sensitivity on two dimensional analysis and 94 percent on three dimensional, against a definition of disease positive as at least one lesion at 50 percent or greater stenosis. No specificity or false positive rate appears anywhere in the company's announcements.

For a triage and notification device that omission is not a technicality, and it inverts in a way worth stating plainly. This product does not add a finding to a report; it reorders a worklist. Every case it flags incorrectly moves some other patient down the queue. So the harm from a false positive here is displacement of a different patient's review, which means the false positive rate is a patient safety number rather than a convenience one. Publishing the sensitivity and withholding the specificity gives a buyer exactly the half that cannot be used to estimate the burden the department will carry.

Ask for specificity, the per study alert rate observed in the clearance dataset, and the composition of that dataset by diabetes status and by site.

One reporting choice here does more work than the headline figure and deserves naming as a general standard. Patient level sensitivity is published for both analysis modes, and the positive case definition is stated explicitly as at least one lesion at a named stenosis threshold.

Publishing the case definition alongside the sensitivity is what makes the number interpretable at all, because sensitivity is entirely a function of what counts as a case: a tool detecting severe disease will report a high figure against a strict definition and a much lower one against a looser one, and the same product can honestly quote either.

Most vendors in this index publish the number without the definition, which leaves a reader unable to tell whether the figure describes finding obvious disease or subtle disease. Held at C because half the operating characteristic is absent.

No specificity is published, and for a triage device that is the more consequential half: a system that reorders a worklist by flagging positives generates a workload cost with every false flag, and a department cannot judge whether triage helps without knowing how often it misdirects attention. Model architecture and validation cohort composition are also unpublished, and no warranty, indemnity or remediation commitment attaches. Ask for specificity at the stated case definition, the positive predictive value in your own case mix, and performance on lesions at lower stenosis grades.

Designed to integrate into existing imaging workflows and radiology worklists rather than requiring a separate application, which is the necessary condition for a triage product to function at all. No named PACS integrations, standards support, or EHR connectivity were located.

No hosting, tenancy or data residency disclosure was located, and for this product type the gap has a specific shape.

A triage and notification device sits in the path between the scanner and the radiologist's worklist, and the decision it returns has to arrive before the study is read to be worth anything. That timing constraint usually pushes vendors toward processing close to the imaging estate, but this company says nothing about it, so a buyer cannot tell whether complete diagnostic CT volumes leave the hospital network on every relevant study or never leave at all. Those are very different procurement propositions and the difference is invisible from the marketing.

The recent listing adds a second question. A British Columbia parent with a Boston operating business raises where processing and storage actually sit, and whether any imaging or derived data crosses the border in either direction. Cross border movement of patient imaging is a question a hospital's privacy office will ask, and it is easier to answer before deployment than after.

Ask for the deployment topology, the processing location, whether an on premise option exists, and what if anything is transmitted outside the customer's network.

No published pricing or commercial model. Worth noting for buyers that the company is publicly listed on a Canadian exchange while very early commercially, which means financial disclosures exist but tell you about the company rather than the cost of the product.

Narrow and precisely bounded: peripheral vascular disease detection on lower extremity CT within radiology workflows, targeted at patients at risk of acute or chronic limb ischemia. Buyers are radiology departments and imaging providers, with vascular surgery and podiatry as downstream referral beneficiaries. The company states a broader vision for AI driven vascular care, which should be read as roadmap rather than available capability.