Medical Imaging · · 5 min read

Making medical imaging systems talk: DICOM, HL7/FHIR, and HIPAA-safe pipelines

How modalities, PACS, and clinical systems actually interoperate — DICOM and DICOMweb, HL7 and FHIR, IHE workflows, de-identification, and building it HIPAA-consciously.

Medical imaging runs on standards that are decades old, deeply entrenched, and unforgiving of shortcuts — for good reason. A CT scanner, a PACS archive, a radiology worklist, and an EHR all have to agree on what a study is, who the patient is, and where the images live. Getting them to interoperate cleanly, without leaking protected health information, is precise work. Here's the shape of it.

EHR HL7 / FHIR Modalities CT · MR · US · X-ray PACS DICOMweb WADO · QIDO · STOW Viewer / AI DICOM de-identify
Imaging (DICOM / DICOMweb) and clinical systems (HL7 / FHIR) meet at the PACS — with a de-identification boundary before research or AI.

DICOM: the language of images

DICOM is the standard for medical images and everything attached to them. It's not just a file format — it's a network protocol and a data model that encodes the image, the acquisition parameters, and the patient and study context together. Modalities (CT, MR, ultrasound, X-ray) produce DICOM; PACS systems store and serve it; viewers consume it.

DICOMweb is the modern, web-friendly face of DICOM — RESTful services for the operations that used to require heavyweight protocols:

  • WADO-RS to retrieve images and metadata
  • QIDO-RS to search for studies and series
  • STOW-RS to store new instances

If you're building anything that touches imaging from a modern application — a web viewer, an AI pipeline, a cloud archive — DICOMweb is usually how you do it without reimplementing legacy plumbing.

HL7 and FHIR: the clinical context

Images don't exist in isolation. Orders, patient demographics, and results flow through the rest of the hospital in HL7 — the long-standing v2 messaging standard that still moves most clinical data — and increasingly FHIR, the modern, resource-oriented, API-friendly standard. A working imaging integration bridges both worlds: the imaging side (DICOM) and the clinical side (HL7/FHIR), so an order becomes a worklist entry, a completed study becomes a result, and the EHR and PACS stay in agreement about reality.

IHE: the workflows that tie it together

Standards define the vocabulary; IHE profiles define how to actually use them together for real clinical workflows — scheduled procedure steps, consistent patient identity across systems, results reconciliation. Aligning to IHE is what keeps a multi-vendor environment coherent instead of a pile of technically conformant systems that still don't cooperate.

PHI is the constraint that shapes everything

Imaging data is protected health information, and that's not a feature you add at the end — it's a constraint that shapes the whole design:

  • De-identification for research, AI training, or any use outside direct care — and done correctly, because DICOM hides identifying data in more places than people expect, including burned into pixels.
  • Access control and audit so every access to a study is authorized and logged.
  • Encryption in transit and at rest, with key management that holds up.
  • Minimal movement. The safest PHI is the PHI you didn't copy. Architectures that keep identified data in one controlled place, and move de-identified derivatives instead, are easier to secure and to defend.

Building HIPAA-consciously means these are designed in from the first diagram, not bolted on before an audit.

IdentifiedPACS · clinical care De-identifystrip PHI + pixel data De-identifiedresearch · AI · sharing
PHI never crosses into research or AI unstripped — de-identification (including burned-in pixel data) is the boundary.

Where the projects get hard

The trouble usually isn't any single standard — it's the seams. A modality that implements DICOM slightly differently. An HL7 feed with local quirks. Patient identity that doesn't quite match across PACS and EHR. De-identification that missed a field. These are the details that decide whether an imaging platform is trustworthy, and they're exactly the kind of thing that's invisible in a demo and expensive in production.

If you're integrating modalities, PACS, and clinical systems — or building a de-identified pipeline to feed imaging AI — that's specialized work where the standards and the safety both have to be right. A scoped assessment is a good way to map the integration and the PHI boundaries before anyone writes the first adapter.