arrow_backRadFuzion field notes
RadFuzion R logoIntegration Architecture10 min read

DICOM router vs HL7 interface engine

A DICOM router moves images. An HL7 interface engine moves the messages about patients, orders and results. Here is what each one does, where they overlap, and how to tell whether you need one or both.

RadFuzion TeamUpdated
Images and messages
One moves studies. The other moves the messages about them.
DICOM router
  1. Modality or PACSC-STORE over DIMSE
  2. Rules on DICOM tagsAE title, first sender, any tag
  3. PACS, archive, cloudC-STORE or STOW-RS
Orders feed the router’s worklist, prior fetching and image holds. Results go back as HL7 ORU.
HL7 interface engine
  1. RIS, EHR, labHL7 v2 over MLLP
  2. Filter and transformFields, codes, segments
  3. Every system that needs itOrders, results, ADT
Simplified. Systems and connections vary by site.
In this article8

What is the difference between a DICOM router and an HL7 interface engine? A DICOM router receives imaging objects over DICOM, decides where each study should go, and delivers it to PACS, archives, workstations or cloud stores. An HL7 interface engine receives HL7 messages, such as orders, results and patient updates, then filters, transforms and delivers them to the systems that need them. The router handles images; the engine handles messages.

The two get confused because they sit in the same place in a diagram, the middle, and because each has grown a little of the other's job. A router that reads HL7 orders can build worklists and fetch priors. An engine can route a message on any field in it. This article separates the two jobs, shows where they overlap, and ends with a checklist for deciding what your workflow needs.

Definitions

Two jobs, two kinds of traffic

A DICOM router is a store-and-forward node for imaging. Modalities and other systems send it studies with C-STORE. It reads each object's header, matches it against rules, can change tags or compression, and forwards the study to one or more destinations. A router should also queue when a destination is down, retry, and keep a record of what went where. Its world is studies, series and instances, identified by UIDs and accession numbers.

An HL7 interface engine is a message broker for clinical systems. The RIS, EHR, lab and billing systems send it HL7 v2 messages, usually over MLLP. The engine parses each message, filters it, maps fields from one system's layout to another's, translates codes, and delivers copies to every system that should see it, each in the format that system expects. Its world is messages: orders (ORM or OMI), results (ORU), patient administration (ADT), scheduling (SIU) and more.

Protocols

DIMSE and MLLP compared

The clearest way to see the difference is on the wire. DICOM network services (DIMSE) start with an association: the sender proposes the kinds of object it wants to send and the encodings it can use, and the receiver accepts or rejects each one. HL7 v2 over MLLP is much simpler. The sender opens a TCP connection, writes a framed text message and waits for an ACK.

How DICOM DIMSE and HL7 v2 over MLLP differ
AspectDICOM (DIMSE)HL7 v2 (MLLP)
What travelsDICOM objects: images, structured reports and other SOP instances, with their headersText messages made of segments such as MSH, PID, ORC, OBR and OBX
ConnectionAn association, negotiated with AE titles and presentation contexts (SOP class plus transfer syntax)A TCP connection; each message is framed by a start byte (0x0B) and an end sequence (0x1C 0x0D)
AcknowledgmentA status on every C-STORE response: 0000 success; Bxxx (and a few 01xx codes) warning; Axxx, Cxxx and most 01xx/02xx codes failureAn ACK message whose MSA-1 is AA (accepted), AE (error) or AR (rejected)
Typical sizeFrom kilobytes to hundreds of megabytes per objectUsually a few kilobytes per message
AddressingHost, port and the called and calling AE titlesHost and port; sending and receiving systems are named in MSH-3 to MSH-6
Where it is definedDICOM PS3.7 (messages) and PS3.8 (network)HL7 Version 2, with MLLP as the usual transport

The two jobs need different machinery. Moving a breast tomosynthesis object of several hundred megabytes needs association handling, transfer-syntax negotiation, compression and long timeouts. Moving a steady stream of small HL7 messages needs parsing, field mapping, ordered queues and quick acknowledgment. The references are the DICOM standard's PS3.7 and PS3.8, and HL7 International's Version 2 product suite and MLLP transport specification.

Orders around routing

What a DICOM router does with orders

An image header says what the modality knew when it sent the study. The HL7 order says what was actually ordered, for whom, how urgently, and whether it has been canceled. A router that receives orders can use that context for the work around each study:

  • Modality worklist. The order becomes a DICOM Modality Worklist entry, so the technologist picks the patient from a list instead of typing demographics.
  • Prior retrieval. The order arriving is the signal to fetch the patient's relevant prior studies from an archive, before the new images exist.
  • Holding images for the order. Images that arrive before their order can wait until it does, matched by accession number, so the PACS does not receive a study ahead of the order it belongs to.
  • Results. A final report that arrives as a DICOM Basic Text SR can go out as an HL7 ORU result.

Note what is missing from that list: choosing the destination. In RadFuzion, routing rules match on DICOM tags and the calling and called AE titles (from 6.8.5, also the AE titles and IP address of the system that first sent the study). They do not read HL7 fields. HL7 orders drive the workflow around routing; the routing decision itself comes from what is in the study. When the header lacks something a rule needs, the fix is to correct the tag in flight with a rule or a lookup table, so the study that reaches the PACS is correct too.

The engine's job

What an interface engine does

An interface engine earns its place when many systems exchange many message types. Its typical work:

  • Fan-out. One order from the EHR goes to the RIS, the PACS, the billing system and a reporting system, each with its own field layout.
  • Transformation. Copying and setting fields, translating codes with lookup tables, reformatting dates, adding or removing segments, splitting report text into OBX lines.
  • Patient administration. ADT feeds keep demographics in step across systems: registrations, updates (A08) and merges (A40). Merges in particular need care, because every downstream system has to apply them the same way.
  • Protocol bridging. MLLP in, HTTP or SFTP or a file drop out, and the reverse.
  • Reliable delivery. A queue per destination, retries, ordered delivery where order matters, and a message log you can search, replay and resend from.

An interface engine is not where images go. Pixel data, associations and transfer syntaxes are a different job, and the engine usually sees imaging through the HL7 messages about it, not the images themselves.

Overlap

Where the two overlap

The overlap is real and it is mostly about orders and results. Both a router and an engine may receive the same ORM. Both may produce an ORU. Both queue, retry and need monitoring. A router can also create HL7: RadFuzion can build an HL7 ORM^O01 from a study's DICOM tags for sites that have no order feed (RadDicom7), and can turn final reports that arrive as DICOM Basic Text SR into HL7 ORU^R01 results with preliminary, final and corrected status.

The risk in the overlap is duplicated logic. If the engine rewrites the procedure code in an order and the router separately rewrites the study description in the images, the two rules will drift apart over time. Decide which system owns each piece of the transformation, and write that down.

Architecture

When you need both

Some common shapes:

An imaging center with one RIS

One RIS sends orders, and a cloud PACS receives the images. A router that consumes the RIS orders directly can run the worklist, prior fetching and image holds. If nothing else needs a copy of those messages, a separate engine may add little.

A hospital with an EHR and many downstream systems

The EHR is the source of orders and ADT, and a dozen systems need them. The engine is already the hub. The router becomes one more subscriber: the engine sends it the orders it needs, in the layout it expects, and receives its results to pass on.

Several sites, several order formats

Each site's RIS puts the accession or the exam code in a slightly different field. Let the engine normalize the messages so the router sees one layout. That keeps the router's configuration the same at every site.

Results that many systems need

The router turns a final report into an ORU. The engine then delivers it to the EHR, the billing system and the patient portal, each formatted the way that system wants.

RadFuzion and RadEngine7

How RadFuzion and RadEngine7 split the work

RadFuzion brings DICOM routing, HL7 order workflows, automated prior retrieval, cloud image management and on-site downtime reading into one platform. On the HL7 side it is a consumer and producer of radiology messages, not a general message broker:

  • It receives HL7 v2 orders over MLLP, on port 2575 by default, or MLLP over TLS on a second port, and acknowledges each message once the order is stored. You choose which fields hold the accession, MRN, exam and the other values it uses.
  • An order can fill the modality worklist (RadList), start prior fetching (RadFetch), appear in the downtime reading list (RadRead) and release images held for it. An order cancel (ORC-1 = CA) removes it from the worklist and stops its prior fetch.
  • Reports go out as HL7 ORU^R01 results: preliminary reads from RadRead over MLLP, and final reports that arrive as DICOM Basic Text SR over MLLP or to Amazon S3 or Mirth. Both are retried automatically.
  • Orders are matched to images by accession number. From RadFuzion 6.8.5, ADT patient updates and merges, and order updates that change the patient, correct exams it has not sent yet. It keeps no master patient index, so reconciling patients across systems stays in your engine, RIS or PACS.

RadEngine7 is the message side. It is a no-code HL7 v2 interface engine that comes with every RadFuzion license and runs as a separate service that pairs with RadFuzion. Its channels receive over MLLP, HTTP, a folder, S3, SFTP or FTP/FTPS, and send over MLLP, HTTP, a folder, S3, SFTP or FTP, or as a PDF. MLLP can run over TLS in both directions, and HTTP listeners and its web page can use HTTPS. Steps such as set, copy, lookup tables, date conversion, find and replace, segment edits and if/else cover common transformations, with sandboxed JavaScript for the rest. Each channel chooses when to acknowledge: on receipt, after delivery or not at all. Alerts go to email, Teams, Slack, Google Chat, a webhook or RadFuzion.

A typical split: the RIS sends orders to RadEngine7, which fixes the fields that vary by site and sends the orders on to RadFuzion and to anything else that needs them. RadFuzion runs the worklist, priors and image holds, routes the studies, and sends results back through RadEngine7 to the EHR.

Limits to plan around

  • MLLP over TLS is off by default and has to be supported at the other end too. Links that stay on plain MLLP belong on a trusted network or a VPN. Certificate revocation (CRL or OCSP) is not checked.
  • RadEngine7 channels have no database, email, DICOM or FHIR connectors (email is for alerts only). If your interfaces depend on those, keep your existing engine and let RadFuzion consume orders from it.
  • RadFuzion matches orders by accession number alone, not by MRN or issuer.

For more on the order workflows, see HL7 and DICOM in one router, and for the routing side, the DICOM router page.

Take it with you

A decision checklist

Answer these for your own environment:

  • How many systems send or need HL7 messages? One RIS feeding one router is a different problem from an EHR feeding a dozen systems.
  • Which message types matter: just orders and results, or ADT, scheduling and documents too?
  • Who owns patient merges and demographic updates, and does every imaging system apply them?
  • Do the images need the order before they are sent on, and how are images matched to orders?
  • Which system owns each transformation: exam codes, study descriptions, accession formats?
  • Does the modality worklist come from your PACS, your RIS or a separate provider?
  • Where do priors live, and what starts their retrieval?
  • Which links need encryption, and how will HL7 and DICOM traffic be protected on the network?
  • Who gets alerted when an interface stops, and who can replay what was missed?

If your answers mostly point at images and orders for radiology, a router that understands orders may cover the job with a small engine beside it. If they point at many systems and many message types, the engine is the hub and the router is one of its spokes. Both designs work. The mistake is letting the two products quietly solve the same problem twice.

See it in your workflow

Put the router and the engine side by side.

RadFuzion routes studies and acts on HL7 orders. RadEngine7, a no-code HL7 v2 interface engine, comes with every RadFuzion license and pairs with it.