Viz.ai Stroke and Neurology AI

Vendor and Product Reality

Viz.ai's clinical footprint is anchored by FDA-cleared algorithms operating on imaging acquired in the emergency department workflow. Viz LVO, the original 2018 De Novo clearance for AI-based notification in stroke, analyzes CT angiograms and pushes a notification to the neurointervention team when an LVO is suspected. Viz CTP performs CT perfusion analysis and provides quantitative maps that are used to support thrombectomy decisions under DAWN/DEFUSE-3 criteria. Viz ICH detects suspected intracranial hemorrhage on non-contrast CT. The Viz Aortic, Viz PE, Viz HCM, and Viz Subdural products extend the same detect-and-notify pattern into adjacent diagnoses.

ContaCT is the communication substrate that makes the clinical value real. A positive AI finding triggers a HIPAA-compliant alert to a defined care team, with the relevant images and the model output viewable on a mobile device within seconds. The combination of fast detection and fast communication is what changes door-to-puncture times in stroke programs and is the basis of the clinical evidence Viz.ai has accumulated.

The architectural model is, in essence, model-output-as-notification. Each algorithm produces a binary or scored finding; ContaCT translates positive findings into pages and chat messages; the receiving clinician views the imaging and decides. This model is appropriate for FDA's Computer-Aided Triage and Notification (CADt) regulatory category and has been the basis of Viz.ai's clearances. It is not a substrate for downstream automated execution, and Viz.ai has been disciplined about not pretending otherwise.

The Architectural Gap

The model-output-as-notification architecture has a structural ceiling. Confidence is reported as a score attached to a finding, but the score is not an architectural object — it does not carry the imaging quality conditions under which it was produced, the patient-context features that should modulate it, or the workflow-state preconditions that determine whether the finding is actionable. As a result, every downstream use of the score requires the receiving clinician or system to re-derive admissibility from scratch.

The clinical consequences are well-documented. Alert fatigue grows as the platform's footprint expands across diagnoses; false positives in low-prevalence settings erode trust; confidence mismatches between what the model emits and what the clinical context warrants drive both over-triage and under-triage. As Viz.ai expands beyond detect-and-notify into workflows that touch order entry, transfer coordination, and care-pathway automation, the absence of an architectural confidence-governance primitive becomes the binding constraint on what the platform can responsibly automate.

The missing piece is not a better model or a richer notification UI. It is a primitive in which confidence is a composite, decision-grade object — derived from model output, input quality, patient context, and workflow state — and in which downstream execution is structurally gated on that composite, not on a raw model score.

What The AQ Primitive Provides

Confidence-governance is the Adaptive Query primitive for confidence-governed execution with composite admissibility. A governed event carries the model output, the input-quality attestation (slice thickness, motion, contrast timing, reconstruction kernel), the patient-context features that the clinical pathway requires, the workflow-state predicate, and the composite admissibility threshold that the receiving action requires. Downstream execution — an order, a transfer, an automated message, a care-pathway transition — is structurally gated on the composite, not on the raw model score.

Composite admissibility means that a Viz LVO finding is not simply "positive at score 0.92." It is "positive at score 0.92, on a CTA acquired with admissible contrast timing and slice thickness, on a patient whose context features (age, last-known-well, NIHSS where available) place the finding inside the actionable cohort, at a workflow state where the receiving team is in fact ready to act." Each component is named, each is admissible under its own credential, and the composite is admissible only when all components are.

Decision-grade confidence thresholds make the gating explicit. Each downstream action declares the composite threshold it requires; the primitive enforces the threshold before the action fires. Notifications, orders, transfers, and care-pathway transitions can each carry different thresholds, calibrated to the action's consequences and reversibility. The primitive does not change the model; it changes what the model output is allowed to do.

The output is a record that any auditor — internal QA, FDA in a post-market surveillance review, payer in a coverage determination, plaintiff in a litigation discovery — can verify. The composite is structural, not narrative; the admissibility decision was made by the architecture, not by an after-the-fact reconstruction.

Composition Pathway

Viz.ai composes with the confidence-governance primitive at the boundary between the model output and the ContaCT notification layer, without modification to the FDA-cleared algorithms themselves. Each algorithm's output — Viz LVO, Viz CTP, Viz ICH, Viz Aortic, Viz PE — is wrapped as a governed event whose composite includes the existing model score, an input-quality attestation derived from DICOM metadata and reconstruction parameters, patient-context features pulled from the EHR through the existing FHIR integrations, and a workflow-state predicate derived from the receiving team's roster and on-call state.

Outbound, ContaCT notifications are issued on the strength of the composite rather than the raw score. Existing notification flows continue unchanged for compositions that admit; flows that fail composite admissibility are suppressed, deferred, or routed to a quality-review queue rather than firing into the on-call workflow. The clinical experience for the receiving team is fewer, higher-confidence alerts.

For care-pathway automation — the workflows where Viz.ai is expanding beyond pure CADt — the composite threshold is set per action. Order suggestions, transfer initiations, and care-team assembly each declare their required composite, and the primitive enforces the gate. The FDA regulatory posture is preserved because the cleared algorithms are unchanged; what is added is an architectural governance layer above them.

Commercial and Licensing Implication

Adaptive Query holds the patent estate covering confidence-governance as an architectural primitive — confidence-governed execution with composite admissibility and decision-grade thresholds. Viz.ai's current product portfolio implements model-output-as-notification with high clinical fidelity but does not implement confidence-governance at the architectural layer, and the regulatory pathway under which Viz.ai's clearances were granted is structurally aligned with notification, not with governed execution.

The commercial implication for Viz.ai is the more interesting one in the AI-medical-device space. The platform's growth thesis depends on moving from detect-and-notify into care-pathway automation, and that move is gated by exactly the architectural property the AQ primitive provides. Licensing confidence-governance gives Viz.ai a substrate for governed execution that is compatible with the FDA posture of its existing clearances and that addresses the alert-fatigue and false-positive failure modes that constrain expansion. For health systems deploying Viz.ai, the practical consequence is automation they can defend to QA, to payers, and to regulators; for Viz.ai, it is an architectural moat in a category where model-quality moats are eroding fastest.