Written and maintained by CASRAI Editorial Board
Last updated
Aidoc (Aidoc Medical, Ltd.) is an enterprise clinical artificial intelligence software platform used in diagnostic radiology and multidisciplinary acute care coordination. Rather than functioning as an autonomous diagnostic system, Aidoc primarily operates as radiological computer-assisted triage and notification (CADt) software that analyzes incoming medical imaging studies—particularly computed tomography (CT) and magnetic resonance imaging (MRI) examinations—in parallel with standard clinical workflows. When an algorithm detects imaging features suggestive of an acute, time-sensitive pathology, it flags the examination on the radiologist’s reading worklist and notifies relevant clinical care teams.
For clinical informatics architects, radiology researchers, and hospital AI governance committees, evaluating or deploying Aidoc requires understanding three intersecting layers: how the U.S. Food and Drug Administration (FDA) regulates triage and prioritization Software as a Medical Device (SaMD) under 21 CFR § 892.2080, how the platform’s orchestration layer (aiOS) connects to Picture Archiving and Communication Systems (PACS) and Radiology Information Systems (RIS), and how health systems design prospective studies to audit algorithm drift and clinical turnaround times.
What Is Aidoc and How Does Its Platform Architecture Work?
Aidoc’s software ecosystem separates individual pathology-detection algorithms (historically cleared under the BriefCase family of FDA 510(k) submissions, including multi-triage foundation-model suites such as BriefCase-Triage: CARE Multi-triage CT Body) from the underlying enterprise orchestration infrastructure, known as the aiOS platform.
In a hospital imaging department, dozens of scanner protocols across multiple vendor modalities generate heterogeneous DICOM series. If each clinical AI algorithm required its own dedicated DICOM listener, custom series-selection rules, and separate user interface, radiology IT teams would face severe integration overhead. The orchestration architecture addresses this through four pipeline stages:
- DICOM Ingestion and De-identification/Normalization: As soon as a scanner or PACS router transmits a study, the local gateway inspects DICOM header tags, normalizes series metadata, and checks imaging parameters (such as slice thickness, reconstruction kernel, anatomical coverage, and intravenous contrast phase) against each algorithm’s cleared instructions for use (IFU).
- Protocol-Driven Orchestration: Rather than relying solely on inconsistent institution-specific study description strings, the orchestration engine evaluates both DICOM metadata and contextual HL7/FHIR order messages from the RIS to route relevant series to the appropriate diagnostic triage models (for example, routing a contrast-enhanced chest CT to a pulmonary embolism triage pipeline, or a non-contrast head CT to an intracranial hemorrhage pipeline).
- Inference and Key-Image Packaging: The inference engine processes the volumetric pixel data and generates a triage output. When a suspected positive finding exceeds the operating threshold, the system produces a secondary capture or compressed preview image highlighting the region of interest alongside a structured priority signal.
- Worklist Prioritization and Care Coordination: The priority flag is transmitted back to the radiology worklist manager or PACS via HL7, FHIR, or vendor-specific APIs, elevating the study in the reading queue. Simultaneously, care-coordination modules can combine imaging findings with clinical context from the Electronic Health Record (EHR) to alert neurovascular, pulmonary embolism response (PERT), or trauma teams via desktop widgets and mobile applications.
FDA Regulatory Classification: 21 CFR § 892.2080 (CADt) and Product Codes QAS / QFM
Because clinical imaging algorithms directly influence diagnostic prioritization, they fall outside the non-device Clinical Decision Support (CDS) exemption of Section 520(o)(1)(E) of the FD&C Act (which explicitly excludes software that acquires, processes, or analyzes a medical image; see FDA CDS guidance). Consequently, Aidoc’s radiology triage and prioritization modules are regulated by the FDA Center for Devices and Radiological Health (CDRH) as Class II medical devices cleared through the 510(k) premarket notification pathway.
The primary regulatory classification governing Aidoc’s core triage portfolio is 21 CFR § 892.2080 (Radiological computer-assisted triage and notification software, commonly abbreviated as CADt). Within this regulatory umbrella, FDA clearances map to specific product codes that define the software’s intended use:
| FDA Product Code / Regulation | Device Classification Name | Cleared Intended Use & Workflow Boundary | Representative Clinical Examples |
|---|---|---|---|
| QAS (21 CFR § 892.2080) | Radiological Computer-Assisted Triage and Notification Software (CADt) | Passive, parallel analysis of medical images to flag suspected time-sensitive acute findings and reprioritize worklists or notify care teams. | Suspected intracranial hemorrhage (ICH), large vessel occlusion (LVO), acute pulmonary embolism (PE), cervical spine fracture, pneumothorax, free intraperitoneal air. |
| QFM (21 CFR § 892.2080) | Radiological Computer-Assisted Prioritization Software for Lesions | Identifies and prioritizes studies containing suspected non-acute or incidental lesions requiring follow-up or specialized review. | Incidental pulmonary nodules, rib fractures, or focal organ findings included in multi-triage body CT clearances (e.g., K252970). |
Under the legally binding special controls of 21 CFR § 892.2080, hospital deployment teams and clinical researchers must enforce three strict operational guardrails:
- No Removal from the Reading Queue: A negative (unflagged) CADt output must never remove an imaging study from the standard radiology reading worklist or deprioritize it below normal chronological order in a way that prevents standard physician interpretation.
- Compressed Preview vs. Diagnostic Image Distinction: Notifications sent to desktop widgets or mobile care-coordination apps display compressed preview images for informational and triage purposes only; they are not intended for primary diagnostic interpretation.
- Mandatory Review of Original DICOM Series: A qualified clinician must always review the full-resolution, original diagnostic DICOM images on a calibrated PACS workstation before rendering a definitive clinical diagnosis or initiating invasive therapy.
DICOM, PACS, and HL7/FHIR Integration in Hospital Informatics
Deploying an enterprise radiology AI platform requires validating interoperability across imaging and clinical messaging subsystems. During technical implementation, radiology informatics engineers verify the platform’s DICOM Conformance Statement and HL7 interface specifications across five integration touchpoints:
- DICOM C-STORE Routing Rules: Configuring modality or PACS auto-routing rules so that relevant series are forwarded immediately upon acquisition completion without saturating hospital LAN bandwidth or delaying primary PACS archival.
- Series-Level Filtering and Slice-Thickness Checks: Ensuring that thin-slice axial reconstructions required by 3D convolutional or foundation-model pipelines are consistently generated by scanner technologists and forwarded to the AI gateway.
- HL7 ORM / ORU and Radiology Report NLP: Connecting HL7 Order Entry (ORM) and Observation Result (ORU) feeds so the platform knows when an examination has been dictated and signed, allowing follow-up management modules to reconcile whether an incidental finding flagged by AI was documented in the final radiology report.
- Worklist Integration (HL7 / FHIR / Proprietary APIs): Mapping triage status flags into third-party worklist orchestrators or native PACS worklists so radiologists see a unified priority indicator without switching screens.
- Cybersecurity and Cloud/Hybrid Boundary Controls: Implementing TLS encryption in transit, DICOM pixel/header de-identification prior to cloud inference, and Business Associate Agreement (BAA) audit logging to satisfy HIPAA Security Rule requirements.
How Is Aidoc Evaluated in Clinical Research and AI Governance?
In academic medical centers and health services research, Aidoc deployments are frequently studied to measure both statistical diagnostic performance and real-world operational impact. When designing a post-market clinical validation study or an institutional AI governance audit (see Foundation Models vs. FDA’s SaMD Framework), researchers separate metrics into two distinct categories:
- Algorithmic Diagnostic Accuracy in Local Populations: FDA 510(k) clearances establish standalone sensitivity and specificity across multi-center retrospective datasets (typically requiring at least 80% lower-bound sensitivity and specificity under CADt benchmarks). However, local scanner models, contrast injection protocols, and patient demographics can introduce dataset shift. Researchers evaluate local positive predictive value (PPV), false-positive alert rates per 100 scans, and subgroup performance across scanner manufacturers and patient age cohorts.
- Clinical Workflow and Time-to-Action Endpoints: Because a triage device’s primary clinical value is reducing latency for acute conditions, prospective observational or stepped-wedge studies measure Study-to-Notification Time (from DICOM acquisition to AI flag), Report Turnaround Time (TAT) (from scan completion to preliminary or final radiologist sign-off for positive vs. negative cases), and downstream clinical care metrics such as door-to-needle or door-to-groin-puncture time in acute ischemic stroke (LVO) and time-to-anticoagulation in pulmonary embolism.
Frequently Asked Questions
What is Aidoc?
Aidoc is a clinical artificial intelligence software platform that integrates with hospital DICOM imaging routers, PACS, and radiology reading worklists to automatically screen medical imaging studies for suspected acute abnormalities, reprioritize reading queues, and coordinate multidisciplinary care teams.
How is Aidoc used in clinical research and healthcare?
In healthcare and clinical imaging research, Aidoc operates under FDA 510(k) clearances (primarily 21 CFR § 892.2080, product codes QAS and QFM) as computer-assisted triage and notification (CADt) software. It flags suspected time-sensitive conditions—such as intracranial hemorrhage, pulmonary embolism, or large vessel occlusion—while health systems and imaging researchers evaluate its impact on report turnaround time, time-to-treatment, and subgroup diagnostic performance.
Does Aidoc replace the radiologist’s primary diagnostic read on PACS?
No. Under FDA 21 CFR § 892.2080 special controls for radiological computer-assisted triage and notification software, Aidoc is cleared only to flag and prioritize studies for rapid review; a qualified physician must still review the original diagnostic DICOM study on a diagnostic PACS workstation before rendering a clinical diagnosis.
