Skip to content
Imaging Evidence Path

What happens after the scanner finishes a study?

Tonight · ~20 min · read · energy: low · setup: none

The scanner produces a study. Then what? This lesson is the journey of an image through a hospital — the actors it meets and the functions each performs — so that every later lesson (how objects move, where AI plugs in, why results vanish) has a map to sit on.

The anchor: functions, not a fixed wiring

Real systems vary wildly: PACS and VNA roles blur or merge, RIS may be folded into the EHR, and the AI service may sit inside or outside the archive. So treat the picture below as a map of functions — archive/store, order/workflow, reporting, viewer/workstation, EHR, AI service — not a universal physical topology. Name the job each actor does, and you can read any site’s variant.

flowchart LR
    M["Modality<br/>(CT/MR)"] --> P["PACS<br/>(store + viewer)"]
    P --> R["RIS<br/>(order/worklist/report)"]
    R --> E["EHR<br/>(clinical record)"]
    P --> V["VNA<br/>(archive role)"]
    P --> A["AI / quant service"]
    A --> D["derived object<br/>(SEG / SR / measurement)"]
    D --> P
    P --> RD["Radiologist reads"]
    RD --> R

The actors, by function

  • Modality — the scanner producing DICOM objects.
  • PACS (Picture Archiving and Communication System) — store, query, distribution, and the viewer/workstation for reading. The radiologist’s daily interface; commonly the primary short/medium-term archive.
  • VNA (Vendor-Neutral Archive) — a standards-based (DICOM) archive role, often for longer-term or cross-site/cross-vendor storage; decouples storage from a specific PACS vendor. Not every site runs a separate VNA.
  • RIS (Radiology Information System) — scheduling, ordering, worklists, reporting, status. Holds the order/accession and report metadata.
  • EHR (Electronic Health Record) — the hospital-wide clinical record; imaging connects to it via results (reports) and increasingly via discrete measurements.
  • AI / quant-analysis service — a consumer of images and producer of derived objects (segmentations, measurements, findings). This is where this path’s work plugs in.

ESTABLISHED EVIDENCE that this architecture exists; COMMON PRACTICE that real deployments blur the PACS/VNA boundary and vary by vendor.

The order, and where things can go wrong

A study is ordered in the RIS/EHR, acquired at the modality (which fetches the order via the worklist), sent to PACS, read by a radiologist at a viewer, and reported back through the RIS to the EHR. An AI service may subscribe to certain studies, retrieve them, run inference, and push a derived object (SEG/SR) back into PACS, where it should appear linked to the source study in the viewer.

Every arrow is a place where identity, provenance, or retrieval can fail — the subject of lesson 6. For now, hold the map: an image is not “delivered” once; it moves through several systems, each of which must reconcile the patient, the study, and any derived results.

Stop and think — then reveal

An AI service segments a tumour and stores the result. The radiologist, reading the study in PACS, never sees the segmentation. Name three informatics (not model) reasons this could happen.

  1. Wrong study attachment — the SEG/SR was stored against the wrong StudyInstanceUID or patient, so it is not linked to the study being read.
  2. Viewer cannot retrieve/display it — the reading workstation does not support the object type (e.g. it cannot render DICOM SEG, or cannot query SR), so the result exists in the archive but is invisible at the viewer.
  3. Lost provenance / no routing rule — no workflow rule routed the AI output to the reading worklist, or the output lacks the source references that would let the viewer associate it with the source series. All three are informatics failures of a correct model output — the subject of lesson 6.

What to retain

  1. Learn the functions (modality, PACS, VNA, RIS, EHR, AI service), not a fixed wiring — real sites blur the roles.
  2. PACS stores + serves + views; RIS owns order/worklist/report; the EHR is where results must ultimately land to affect care.
  3. An AI service is a consumer of images and a producer of derived objects that must re-enter the same system, linked to the source study.
  4. Every arrow is an identity/provenance/retrieval failure waiting to happen — the map sets up the failure-mode lens of lesson 6.

Next: the protocol that actually moves objects between these actors — how DICOM objects move.