AI Radiology Triage Software for a Mid-Size US Hospital: Pilot Aidoc and RapidAI, Not Yet Adopt

Verdict

Interpretation: The category is AI radiology triage (computer-aided triage and notification) for a mid-size US hospital under FDA and HIPAA. I scored Aidoc in full because it has the most recent, verifiable FDA activity and a documented Epic listing. Viz.ai and RapidAI are scored in summary. The must-haves are Epic EHR integration, SOC 2 Type II and an on-prem option. All three were stated by the requester, so none is inferred. Today is 6 October 2026.

Verdict: pilot, with conditions. Aidoc is the strongest fit on FDA clearance and Epic integration. No vendor I reviewed publicly proves all three must-haves. Aidoc and Viz.ai appear to run AI inference in the cloud, so on-prem fails or is unproven for them. RapidAI claims an on-prem option, but I found no Epic listing for it. Confirm these gaps in writing before signing anything. This is a recommendation for your team to confirm, not a purchasing decision, and it says nothing about whether your hospital is compliant.

Scorecard (Aidoc, primary candidate)

Dimension	Finding	Evidence type	Source link	Rating
Regulatory / compliance	FDA 510(k) K252970, BriefCase-Triage: CARE Multi-triage CT Body, received 17 Sep 2025, decision 7 Jan 2026. A second entry, K253578 (pneumothorax, pericardial effusion, large aortic aneurysm, shoulder fracture or dislocation), shows decision 26 Feb 2026, Substantially Equivalent, product code QAS.	Independent (FDA registry)	FDA 510(k) database	Strong
Validation evidence	Vendor reports that in the FDA-reviewed pivotal study the 11 new indications had mean sensitivity of 97% and mean specificity of 98%. I did not retrieve the FDA summary PDF or an independent peer-reviewed study of the CARE model.	Vendor	Aidoc press release (date not shown on fetched page)	Weak
Performance	The same release claims roughly an order-of-magnitude reduction in false alerts against single-condition tools. This is a marketing claim with no independent benchmark found.	Vendor	Aidoc press release	Unknown
Security & privacy	The trust center portal lists HIPAA, ISO/IEC 27001, SOC 2, TX-RAMP and ISO 13485 but the reports are gated. A third-party aggregator notes the SOC 2 type (1 or 2) is not stated on any public page. The portal also logs a 4 Feb 2026 incident involving a third-party website widget: containment the same day and a forensic review of HubSpot and Salesforce integrations. The vendor states no PHI, patient data or customer data was affected and classifies it as a false positive (vendor claim).	Vendor (plus third-party aggregator)	Aidoc trust center	Adequate
Integration	Live in the Epic App Orchard (announcement undated on the page; circa 2022), and the platform connects over DICOM, HL7 and FHIR. The Epic evidence is more than two years old.	Vendor	Aidoc press release; integrations page	Adequate
Pricing & contract	No public pricing found. I did not retrieve Aidoc's terms of service.	None	n/a	Unknown
Vendor viability	$150 million round (July 2025 per my notes; date not on the fetched page), total funding $370 million, led by General Catalyst and Square Peg, with four US health systems participating. Vendor claims 150+ health systems and 1,600+ hospitals.	Vendor	Aidoc funding release	Strong
Risks	Cloud inference, gated SOC 2 detail, undated VA security document, early-2026 website incident, unpriced contract. See the risk list below.	Mixed	n/a	Adequate

Must-have checklist

Requirement	Status	Evidence and source
Integrates with Epic EHR	Pass for Aidoc (dated); unclear for Viz.ai and RapidAI	Aidoc: Epic App Orchard go-live announcement, undated, circa 2022. Viz.ai: a third-party review says Epic integration is institution-specific and not a standard feature, with no native write-back; this is unaudited. RapidAI: I found only a claim of vendor-agnostic integration, no Epic listing.
SOC 2 Type II	Vendor-claimed pass for Viz.ai (Type II stated only in a 2024 release); unclear for Aidoc; vendor-claimed for RapidAI	Viz.ai: trust center lists SOC 2 and HIPAA, with the Type II label not shown there, and a Feb 2024 release reports a Big Four audit. I have not seen the report itself. Aidoc: SOC 2 listed, type not public (see above). RapidAI: lists "SOC2 Type 2" among its certifications, a vendor claim.
On-prem deployment option	Fail (as documented) for Aidoc and Viz.ai; claimed for RapidAI	Aidoc: a VA-hosted security overview (undated, version 1.1) says studies are de-identified and uploaded to the cloud analysis server. Only the orchestration step runs on-premises. Viz.ai: no on-prem option found; a third-party review describes routing studies to the Viz.ai cloud. RapidAI: Hybrid, On-Prem and Cloud models, vendor claim.

Vendor comparison

Vendor	FDA position	Security attestation	Deployment	Viability signal	Source
Aidoc	Two 2026 CARE triage 510(k) decisions (above)	SOC 2 listed, type gated	Cloud analysis, on-prem orchestrator	$370M total funding	Aidoc funding release
Viz.ai	Vendor says 13 FDA clearances (July 2026); clearance numbers not checked by me	SOC 2 Type II, ISO 42001 stated	Cloud (third-party review; vendor subprocessor list is gated)	Headcount 374 in March 2026, down from 528 in 2023 (Revelio estimate); last disclosed $100M Series D at $1.2B valuation in 2022	Viz.ai trust center
RapidAI	Not checked by me	SOC 2 Type 2 claimed, trust center gated	Hybrid, on-prem, cloud (vendor)	Vendor claims 2,500+ hospitals	RapidAI infrastructure page

The Viz.ai headcount decline is a third-party estimate and does not by itself show financial distress. A third-party site also claims Viz.ai has "50+ FDA clearances", which conflicts with the vendor's own count of 13, so treat that site's figures with caution. It also quotes a price range of USD 150,000 to 300,000 per site per year for stroke deployments, which the site itself says is not independently audited. I treat it as an unverified estimate for budgeting only.

Alternatives worth comparing

Viz.ai: strongest published SOC 2 Type II position but cloud only and Epic depth unclear.
RapidAI: the only one of the three that claims on-prem deployment; verify its radiology (non-stroke) triage clearances and Epic fit.
Annalise.ai (Annalise Enterprise, K250831): appeared in an FDA 510(k) listing but I did not assess it; shortlist only if chest X-ray triage is a priority.

Do next

Send each shortlisted vendor the five questions above, in writing, and ask for the SOC 2 Type II report, BAA and subprocessor list.
Pull the FDA summary PDFs for K252970 and K253578, and confirm which clearances cover the exam types your hospital reads.
Ask your Epic team to check each vendor's Showroom listing and your Epic version.
Define the on-prem requirement: is it data residency, or only outage resilience? RapidAI's "on-prem fallback" and Aidoc's on-prem orchestration may satisfy the second.
Run a time-limited pilot on one modality with your own retrospective cases to measure sensitivity, specificity and alert burden.

Open items

Not verified: FDA clearance numbers for Viz.ai and RapidAI; independent peer-reviewed validation of Aidoc CARE; recall, MAUDE, warning-letter and breach-portal checks for all three vendors.
Near-miss evidence: the aggregator aifoxx.com (last verified 29 June 2026) says Aidoc's SOC 2 type could not be confirmed. The VA security overview is undated, so its architecture may have changed.
Pricing: no vendor price was verified from a primary source.

Coverage

Not reached: gated trust-center documents (Aidoc SafeBase, RapidAI, Viz.ai SOC 2 report), the Epic Showroom listing pages, vendor terms of service, G2 and Capterra reviews, and FDA recall and MAUDE databases.