Aidoc Medical Imaging AI
by Nick Clark | Published April 25, 2026
Aidoc operates one of the largest commercial medical-imaging AI deployments, with an FDA-cleared product portfolio spanning intracranial hemorrhage, large-vessel occlusion, pulmonary embolism, cervical-spine fracture, and abdominal pathology detection. The architectural primitive Aidoc lacks — confidence-governed execution with composite admissibility, decision-grade thresholds, and pause/reassess/resume semantics — is exactly what the confidence-governance substrate provides, and it is increasingly the regulatory difference between a notification engine and a decision-grade clinical instrument.
Vendor and Product Reality
Aidoc's commercial footprint is dominated by its aiOS (AI Operating System) platform, which orchestrates a portfolio of FDA-cleared modules including BriefCase for triage of ICH on non-contrast CT, LVO on CT angiography, PE on contrast-enhanced chest CT, and incidental findings such as abdominal aortic aneurysm and pulmonary nodules. Each module operates as an always-on inference layer that intercepts DICOM studies as they arrive at the PACS, runs the relevant detector, and pushes a worklist priority change or a notification to the radiologist's reading environment. The platform integrates with Epic, Cerner, Change Healthcare, Sectra, and most major RIS/PACS vendors through HL7 and DICOM SR pathways.
Aidoc's product strategy has shifted from single-pathology classifiers toward what the company calls "Always-On AI" — a workflow layer that continuously executes its full module suite against every eligible study. Recent additions include the CARE platform, which extends Aidoc's reach into longitudinal patient management, and partnerships with GE HealthCare, Imbio, and Subtle Medical that allow third-party algorithms to ride the same orchestration substrate. The company has more FDA 510(k) clearances than any other radiology-AI vendor and reports deployments across more than one thousand hospitals globally, including the Mayo Clinic, Cedars-Sinai, and a substantial NHS footprint.
Architectural Gap
Aidoc's modules emit per-study probabilities, but the platform itself does not model decision-grade confidence as a first-class governance object. A BriefCase ICH detector will return a positive flag whenever its internal score crosses a fixed operating point, and the downstream workflow treats that flag as binary — either the study is escalated or it is not. There is no native mechanism for the platform to admit that a result is below decision-grade threshold, request additional reconstruction or sequence acquisition, and then resume on the resulting evidence. Composite admissibility — the requirement that a clinical inference be admitted only when corroborated by an explicit set of structural, contextual, and historical signals — is not part of the data contract Aidoc surfaces to ordering clinicians or to the EHR.
The gap matters in two regulatory directions. The FDA's Predetermined Change Control Plan (PCCP) framework, which Aidoc has publicly embraced, contemplates ongoing model updates within a pre-authorized envelope; but PCCP presumes the sponsor can demonstrate that admissibility criteria for inference outputs are stable even as the underlying model drifts. Aidoc today demonstrates this through retrospective performance studies, not through a runtime governance object. Second, payers and health systems increasingly demand audit trails that show why a specific case was escalated and why others were not — a question that requires confidence as a governed, signed runtime artifact rather than a probability buried in inference logs.
What the AQ Primitive Provides
Confidence-governance, as defined in the Adaptive Query architecture, treats every inference as a candidate that must clear composite admissibility before it can affect downstream state. Admissibility is not a single threshold; it is a structured predicate composed of model confidence, corroborating evidence, provenance integrity, and contextual constraints. When admissibility fails, execution does not silently proceed with a low-confidence result — it pauses, surfaces the deficit, and either invokes a reassessment path (such as requesting an additional acquisition or invoking a second-opinion model) or hands control back with an explicit rationale.
The substrate is decision-grade by construction: every admitted inference carries a signed lineage that names the contributing models, the evidence they consumed, the thresholds they cleared, and the reassessment branches that were considered and rejected. Pause/reassess/resume is a runtime primitive, not an application-level retry loop, which means the same governance object is observable to regulators, payers, and downstream automation alike. Critically, admissibility is composable across modules — an ICH detector and a c-spine fracture detector running on the same trauma CT can contribute to a joint admissibility decision rather than emitting independent flags.
Composition Pathway
Integration with Aidoc's existing aiOS does not require replacing any cleared module. The composition pathway treats each Aidoc detector as a confidence-bearing inference source whose outputs are admitted through the governance substrate before they reach the worklist or the EHR. Aidoc's existing DICOM SR outputs and aiOS event stream are wrapped as governed observations; the substrate then evaluates composite admissibility — combining the detector's score with corroborating signals from prior imaging, the indication captured at order entry, and any concurrently running modules — and emits a single admitted finding rather than a raw notification.
For Aidoc's third-party algorithm partners, the same pathway provides cross-vendor admissibility without forcing any partner to expose proprietary model internals. A Subtle Medical reconstruction, an Imbio quantification, and an Aidoc detector can each contribute signed observations that the substrate composes into a decision-grade output. The pause/reassess/resume primitive maps directly onto Aidoc's emerging interest in agentic radiology workflows, where an inference that fails admissibility can request a thin-slice reconstruction, a delayed-phase acquisition, or a specialist consult and then resume on the augmented evidence.
Commercial Implication
Aidoc's commercial conversation with health-system CMIOs has shifted from "does the algorithm work" to "can we trust it to drive workflow." Confidence-governance reframes that conversation: instead of negotiating per-module sensitivity and specificity, the health system contracts for an admissibility envelope that holds across modules and across model updates. This aligns naturally with value-based contracts, where Aidoc's revenue is tied to outcomes rather than per-study fees, because admissibility provides the auditable ledger payers require to validate outcome attribution.
For Aidoc's competitive position against Rapid AI, Viz.ai, and the emerging GE HealthCare and Siemens Healthineers in-house platforms, the substrate provides a defensible architectural moat. Competitors can match individual clearances; they cannot match a governance layer that makes Aidoc's broader portfolio behave as a single decision-grade instrument. The substrate also creates a natural acquisition pathway for smaller specialty-AI vendors — neuroimaging, cardiac, oncology — because each can plug in as a composite-admissibility contributor without bespoke integration work.
Licensing Implication
The confidence-governance primitive is not a feature Aidoc would build internally without a multi-year detour from its core radiology mission. Licensing the substrate gives Aidoc immediate access to PCCP-aligned admissibility semantics, signed lineage suitable for FDA post-market surveillance, and a runtime governance object that satisfies the European AI Act's high-risk-system documentation obligations. The licensing structure contemplates that Aidoc retains exclusive control over its cleared modules and clinical claims while operating on top of a substrate whose admissibility logic is independently maintained and independently auditable.
For Adaptive Query, the Aidoc relationship establishes confidence-governance as the canonical substrate for FDA-regulated inference at the workflow tier — a position that compounds across every subsequent radiology, pathology, and cardiology vendor. The licensing implication is therefore reciprocal: Aidoc gains the architectural element that converts its clearance portfolio into a decision-grade instrument, and the substrate gains the commercial validation that makes it the default governance layer for clinical AI.
