Viz.ai

Viz.ai sells Viz One as an AI powered care coordination platform. Detection models sit under each suite, with more than 50 FDA cleared algorithms across CT, ECG and echocardiography for neurovascular, vascular, cardiac, pulmonary, trauma and radiology work. They are the engine of the alerting module: nothing reaches a specialist's phone until a model flags the study.

The platform also carries value without the models. The mobile app gives clinicians secure image viewing, messaging and team activation across hub and spoke networks, and Viz Assist is sold as a separate workflow assistant. Remove the algorithms and a hospital would still have a care coordination and communication system, so the models are a core module rather than the whole product.

For the original stroke product the oversight model is published in full. FDA's De Novo decision summary for ContaCT (DEN170073, February 2018) describes a notification only tool. It runs in parallel with normal reading and alerts a specialist's phone with the images attached; its findings are not for diagnostic use beyond notification. It flags a study at a stated fixed threshold, and the radiologist's report still follows the standard of care.

That description covers one algorithm, first cleared in 2018. The platform now sells more than 50, and no equivalent statement of limits, thresholds and who sees the unflagged studies is published per module. The homepage also describes "automated assessments to accelerate diagnosis", which reaches further than a notification only device.

The approach is named, the detail is not. FDA's De Novo summary for ContaCT (DEN170073) describes an artificial intelligence algorithm that analyzes CT angiography for large vessel occlusion. It was tested on 300 studies from two US sites: sensitivity 87.8 percent, specificity 89.6 percent, AUC 0.91, and median time to notification of 5.6 minutes against 51.5 minutes for standard care. Later work names the network type for some modules; the hemorrhage quantification tool is described as a convolutional neural network.

The company publishes no architectures or training data and does not say how updates to each of the 50 plus algorithms are released and announced. The 2018 figures describe the first version of one product, not the current release.

The imaging model layer is Viz.ai's own. Each algorithm is FDA cleared per indication, so it cannot be swapped without new review, and no outside model provider is named in the path. The host is named on the Trust Center: Viz.ai uses Amazon, with ISO 27017 and 27018 for cloud controls. A subprocessor list exists but sits behind a request on the Drata trust portal rather than in public.

Two parts of the chain are open. Viz Assist, sold as a workflow assistant, is not described, so whether it relies on an outside language model is unknown. The mobile app puts images on clinicians' own devices, outside the vendor's infrastructure.

The stroke evidence measures care delivery in prospective, multi site designs run by hospital investigators. A post hoc analysis of a multicenter, prospective, randomized trial of Viz LVO, led at UTHealth Houston, reported an 11 minute reduction in door to groin time across all centers. A parallel cohort of acute stroke consults at 166 facilities in 17 states compared AI and non AI sites. Single center pre and post studies report shorter door in to puncture times and higher reperfusion rates after deployment. Viz.ai hosts a publications library on its site.

The strength is concentrated in large vessel occlusion. Most studies are before and after comparisons in which other workflow changes also happened. The evidence for the newer cardiac, pulmonary and oncology modules is thinner, and some are sold as pilot programs.

Viz.ai's privacy policy for its subscription services (effective August 2025) makes categorical commitments. Protected health information in the service is governed by the customer's HIPAA Business Associate Agreement, not the policy. The customer's contract sets how long Viz.ai keeps customer data. Usage data (which studies and alerts a clinician opened, response times) is de-identified to the HIPAA standard at 45 CFR 164.514 before it is used to improve services, build new products or conduct research.

What is not published: a retention schedule, whether patient imaging is used to train or retrain models, and what persists on clinicians' phones after a case closes, which matters because the product pushes imaging to mobile devices by design. No guardrails or safety event process is published for the algorithms.

Business associate status is stated in a privacy document that covers the service. The US privacy policy for Viz.ai's subscription services says protected health information is governed by the HIPAA Business Associate Agreement between the customer and Viz.ai. The Trust Center describes a HIPAA program with administrative, technical and physical safeguards, regular risk assessments including third party evaluations, and SOC 2 controls mapped to HIPAA. HIPAA compliance documentation is available on request through the Drata trust portal.

The agreement itself is not published, so its terms on subcontractors, breach notice and data return cannot be read before contracting. Separate privacy policies cover EU platform users.

The Trust Center linked from Viz.ai's site footer names the company's own certifications. SOC 2 Type II covers security, availability, confidentiality, processing integrity and privacy, scoped to production systems and supporting infrastructure. ISO 27001 is held with ISO 27701, 27799, 27017 and 27018, 22301 and 27035. Controls are monitored continuously through Drata between audits.

The artifacts are not open to read. The SOC 2 Type II report, ISO certificates, the independent penetration testing summary, the subprocessor list and continuity plans are available on request through the Drata portal at trustcenter.viz.ai. Audit periods and certificate dates are not shown publicly.

ContaCT, the company's first product, received FDA's De Novo authorization DEN170073 in February 2018. That created the regulation for radiological computer aided triage and notification software (21 CFR 892.2080, product code QAS), under which many later imaging triage products have been cleared. The company now reports more than 50 FDA cleared algorithms across stroke, hemorrhage, aneurysm, pulmonary embolism, aortic disease and cardiac indications, and publishes an Indications for Use page. EU user guides are published separately for CE marked products.

Clearance is per algorithm and per indication. Some items on the product menu are sold as pilot programs (oncology, respiratory, osteoporosis), and a pilot is not a clearance.

Viz.ai names ISO/IEC 42001, the AI management system standard, among its certifications on the Trust Center. It says the certificate covers how the company develops, evaluates and deploys AI systems. That is a governance framework with an outside audit behind it, and ISO 27701 privacy management sits alongside it. The certificate is available on request through the trust portal, not published.

No bias evaluation or subgroup performance is published for any algorithm. The allocation question is specific here: the product speeds transfers from spoke hospitals to thrombectomy centers, so its benefit reaches only the catchments whose hospitals run it.

An error rate exists for the first product. FDA's De Novo decision summary for ContaCT (DEN170073, 2018) reports a standalone test on 300 CT angiography studies from two US sites against neuroradiologist ground truth. Sensitivity was 87.8 percent (95% CI 81.2 to 92.5) and specificity 89.6 percent (83.7 to 93.9). The FDA wrote that summary, and Viz.ai does not publish or cite the figures itself. They describe a version first cleared in 2018, not the current large vessel occlusion product.

No warranty, indemnity or remedy is attached to a missed or false alert. The missed activation is the harder failure: an unflagged case proceeds at ordinary speed, and nothing records that the system should have caught it.

Integration is imaging and communication. The platform receives DICOM studies from scanners and PACS, and the De Novo summary describes the notification path to the mobile app with the images attached. The app is the working surface for neurologists, interventionalists and stroke teams across hospital networks. The platform runs at a named scale of about 2,000 hospitals and health systems in the United States and Europe, and a named HCA division CIO speaks for it on the Trust Center.

No EHR integration is named or documented, and no PACS or RIS product list is published. Viz.ai lists a Microsoft Cloud for Healthcare collaboration, not read here.

The platform is cloud hosted. The Trust Center says Viz.ai uses Amazon, encrypts data in transit and at rest, and holds ISO 27017 and 27018 for its cloud infrastructure. It processes data across hospitals in the United States and Europe, with EU to US transfers under the EU-US Data Privacy Framework. The design sends imaging to clinicians' phones.

No on premises option is described, and residency is not committed: the region a given customer's data rests in, and whether European data stays in Europe, are not stated. Tenant isolation is not described. The mobile endpoint, meaning what is cached on a phone, for how long, and what happens when a device is lost or a clinician leaves, is not described.

No price is published. The website sets out how the product is sold: Viz One is an enterprise platform with therapeutic suites (neuro, vascular, cardio, pulmonary, trauma, radiology, oncology) and pilot programs, and the route in is a contact form. So cost depends on the suites and hospitals covered, while the unit of charge and any figure go unstated.

Medicare's New Technology Add on Payment has applied to AI stroke triage in past years. That is what a hospital could recover, not what Viz.ai charges, and its status for the current year is not stated.

Coverage is named broadly: stroke and neurovascular care, hemorrhage and aneurysm, pulmonary embolism and aortic disease, cardiac conditions including hypertrophic cardiomyopathy on ECG, trauma, and radiology. It spans CT, ECG and echocardiography, in hub and spoke hospital networks in the United States and Europe, at about 2,000 hospitals and health systems.

Validation stands behind part of it. Large vessel occlusion has a De Novo filing with a two site test set and prospective multi site workflow studies. Other modules have clearances and some studies. The oncology, respiratory, neurology and osteoporosis offers are sold as pilot programs.