Common DICOM routing failures and how to fix them
Routing problems tend to fall into a handful of patterns. For each one: the symptom you see, the likely cause, how to confirm it, and the fix.
- 01NetworkPort, firewall, DNS
- 02AssociationAE titles, presentation contexts
- 03StorageC-STORE status, SOP class, disk
- 04DataUIDs, accession, transfer syntax
- 05CloudTokens, roles, per-instance results
In this article11
Why did a DICOM study not arrive? The cause is usually one of a few failures: the association was rejected, the receiver refused the object, a large transfer timed out, the data itself was wrong, the cloud store refused the upload, or the destination was down. This guide takes each one in turn, with the symptom, the likely cause, how to confirm it and how to fix it.
The advice is vendor-neutral. The last section covers what RadFuzion does about each failure.
Method
Start with where it failed
Before changing anything, find the layer where the transfer stopped. A ping or an open port proves the network. A C-ECHO proves the association and AE titles. A C-STORE status proves the receiver accepted that object. After that, look at the data. For cloud stores, the HTTP status and the response body play the same roles.
Three habits save time:
- Get the exact error from the sender’s log: the reject reason, status code or HTTP response, not just "failed".
- Find out whether it is every study, one modality, one object type or one destination. The pattern points at the layer.
- Reproduce with one known study, ideally a small uncompressed one, before testing with the study that failed.
The codes below come from the DICOM standard: association reject reasons in PS3.8 section 9.3.4, C-STORE statuses in PS3.4 section B.2.3, general DIMSE statuses in PS3.7 Annex C, and STOW-RS responses in PS3.18 section 10.5.3.
Association
Association rejected
Every DIMSE transfer starts with an association. When it fails, the sender usually reports "association rejected" or "no accepted presentation context". The reject reason in an A-ASSOCIATE-RJ, or the result for each presentation context in an accepted association (DICOM PS3.8, section 9.3), tells you which of three problems you have.
Calling or called AE title not recognized
- Symptom
- Every send to one destination fails at once, including C-ECHO. Other senders to the same destination work.
- Likely cause
- The destination does not know the sender’s calling AE title, or the sender is using the wrong called AE title for the destination. Many systems compare AE titles case-sensitively.
- How to confirm
- Read the reject reason: 3 is calling-AE-title-not-recognized and 7 is called-AE-title-not-recognized. Compare the exact spelling and case on both sides.
- Fix
- Add the sender’s AE title, IP address and port to the destination’s list of known nodes, or correct the called AE title on the sender. Retest with C-ECHO.
Unsupported SOP class or transfer syntax
- Symptom
- The association opens, but some studies, or some series within a study, do not arrive. Often it is one object type: a structured report, a secondary capture or an ultrasound clip.
- Likely cause
- The receiver accepted the association but rejected the presentation context for that object: abstract-syntax-not-supported (result 3) or transfer-syntaxes-not-supported (result 4).
- How to confirm
- Check the sender’s log for the object’s SOP Class UID and transfer syntax, and the receiver’s DICOM conformance statement for what it accepts.
- Fix
- Enable the SOP class on the receiver, or offer an uncompressed transfer syntax (Explicit or Implicit VR Little Endian) alongside the compressed one. If the receiver cannot store that SOP class at all, route those objects elsewhere.
Too many associations
- Symptom
- Sends fail intermittently at busy times and succeed when retried.
- Likely cause
- The receiver has a limit on simultaneous associations, and the senders together exceed it. The reject reason is often temporary-congestion or local-limit-exceeded, though some systems simply refuse the TCP connection.
- How to confirm
- Count concurrent associations from every sender during the failure window, and compare with the receiver’s configured limit.
- Fix
- Lower the number of parallel sends on the sender side, or raise the limit on the receiver if it has capacity.
Storage
C-ECHO works but C-STORE fails
A successful C-ECHO proves that the network path, the port and the AE titles are right. It does not prove that the receiver will store anything: verification is its own SOP class.
C-STORE returns a failure status
- Symptom
- Echo succeeds, but every C-STORE, or every C-STORE of one kind, is refused.
- Likely cause
- The storage presentation context was rejected, or the receiver answered with a failure status: A7xx (out of resources, often a full disk), A9xx (data set does not match the SOP class) or Cxxx (cannot understand). 0122, a general DIMSE status, means the SOP class is not supported.
- How to confirm
- Find the status code in the sender’s log. C-STORE statuses are listed in DICOM PS3.4 section B.2.3, and general DIMSE statuses in PS3.7 Annex C.
- Fix
- For A7xx, free space or fix the receiver’s storage. For A9xx and Cxxx, look at the object itself: missing required attributes or a malformed header are common. For 0122, enable the SOP class or send the object elsewhere.
Size
Timeouts on large studies
Big studies fail, small ones succeed
- Symptom
- Tomosynthesis, large CT or whole-slide studies fail part-way, while small studies to the same destination go through.
- Likely cause
- A send timeout sized for small objects, a firewall or load balancer that drops long-lived idle connections, or a slow WAN link. Batching many large objects into one association makes it worse.
- How to confirm
- Compare the failure time with the configured timeout. If transfers consistently stop after the same number of minutes, suspect a timeout on the sender, the receiver or a device in between.
- Fix
- Raise the send timeout for large modalities, send large objects one per association, and check firewall session timeouts.
Compression
Compression and transfer-syntax problems
Images arrive but do not display, or conversion fails
- Symptom
- The destination stores the study, but a viewer shows blank or garbled images, or a conversion step reports a codec error.
- Likely cause
- The destination accepted a compressed transfer syntax that a downstream viewer cannot decode, or a converter lacks the codec for the source encoding. Converting to JPEG Baseline is lossy, which may not be acceptable for the archive of record.
- How to confirm
- Note the Transfer Syntax UID (0002,0010) of the stored object and check it against what each reader supports.
- Fix
- Send in a syntax every reader supports: uncompressed, or a lossless syntax such as JPEG 2000 Lossless. Keep lossy compression for copies that are not the archive of record, and make that a deliberate decision.
Identity
Duplicates, missing identifiers and split studies
Duplicate or overwritten instances
- Symptom
- A corrected study is resent and the destination still shows the old values, or shows two copies.
- Likely cause
- The resent objects have the same SOP Instance UIDs. Receivers handle that differently: some replace the stored object, some keep the original and report it as already stored, some reject it.
- How to confirm
- Compare SOP Instance UIDs between the copies, and check the receiver’s duplicate policy.
- Fix
- Know each destination’s policy before you correct data. Where the destination keeps the original, delete the stored study, then resend.
Missing accession number or UIDs
- Symptom
- Studies land in the PACS unmatched to an order, or are refused on receipt.
- Likely cause
- A modality not using the worklist, a manual entry, or an outside study. A missing Study, Series or SOP Instance UID makes the object invalid.
- How to confirm
- Check Accession Number (0008,0050) and the three UIDs in the received objects.
- Fix
- Put the modality on the worklist so the accession comes from the order. Hold studies missing required tags for a fix instead of forwarding them, or generate an accession by rule where the workflow allows it.
Studies split or merged wrongly
- Symptom
- One exam appears as two studies in the PACS, or two exams appear as one.
- Likely cause
- Late images or post-processing results arrive with a new Study Instance UID, or the router decides a study is complete too early and sends the late images as a separate delivery. In the other direction, two procedures under one accession get merged.
- How to confirm
- Compare Study Instance UIDs, accession numbers and arrival times for the affected images.
- Fix
- Keep late images on the original Study UID for that accession, give the completion window enough time for the modality’s real behavior, and treat one accession with several procedures as its own case.
DICOMweb
Cloud STOW-RS errors
Cloud DICOM stores take images over STOW-RS, an HTTPS POST. The HTTP status describes the request; the response body lists the outcome for each instance (DICOM PS3.18, section 10.5.3).
401 or 403
- Symptom
- Every upload to one store fails with 401 Unauthorized or 403 Forbidden.
- Likely cause
- 401: the access token is missing, expired, or issued for the wrong audience. 403: the identity is valid but lacks permission to write to that store.
- How to confirm
- For 401, request a token by hand with the same credentials. For 403, check the role on the store: on Google Cloud, a role that allows DICOMweb writes, such as Healthcare DICOM Editor; on Azure, DICOM Data Owner.
- Fix
- Correct the credentials or the role. Make sure the sender requests a fresh token when one expires during a long upload.
409 Conflict, or 202 Accepted with failures
- Symptom
- The upload "succeeds" but images are missing, or a resend reports a conflict.
- Likely cause
- 202 means some instances were stored and others had warnings or failures. 409 means none were stored: often because they already exist, sometimes because of an unsupported SOP class or a Study UID mismatch.
- How to confirm
- Read the Failed SOP Sequence (0008,1198) in the response body. Each entry gives the instance and a failure reason.
- Fix
- Count delivery per instance, not per request. For "already exists", decide whether the stored copy is correct; if it is not, delete it and resend.
413, or timeouts on large uploads
- Symptom
- Large studies fail with 413 Payload Too Large, a gateway timeout, or a dropped connection.
- Likely cause
- Too many instances in one multipart request, a proxy body-size limit, or a request that runs past the server’s or a proxy’s time limit.
- How to confirm
- Compare the request size and duration with the limits of the store and any proxy in the path.
- Fix
- Send fewer instances per request, one for very large objects, and raise proxy limits where you control them.
Availability
Destination down and backlogs
A destination stops answering
- Symptom
- Studies pile up on the router, or on the modalities if the router itself is down.
- Likely cause
- The destination, its network or its DNS is unavailable, or a certificate or credential expired.
- How to confirm
- Run C-ECHO or a test upload from the router, and check the destination’s own status.
- Fix
- Hold studies on the router and replay them when the destination returns. Watch free disk on the router during long outages, and allow time for the backlog to drain.
Orders
Hold-for-order timeouts
Images wait for an order that does not match
- Symptom
- Studies sit in a hold queue, then time out, even though the order was sent.
- Likely cause
- The accession in the HL7 order differs from the one in the images (leading zeros, a prefix, a different field), the order interface is down, or the order arrives later than the timeout.
- How to confirm
- Search the order log for the study’s accession, and compare the exact value in the order with Accession Number in the images.
- Fix
- Map the accession from the field the RIS actually fills, make the two formats agree, and set the timeout to cover the real gap between images and orders.
Time
Clock and time-zone problems in schedules
After-hours rules fire at the wrong time
- Symptom
- Overnight or weekend routing, for example to a teleradiology group, starts an hour early or late, or skips the hours after midnight.
- Likely cause
- The rule is evaluated on a server clock in another time zone (cloud servers often run on UTC), a daylight-saving change, or a window that crosses midnight written as two separate rules. Wrong modality clocks also skew Study Date and Time.
- How to confirm
- Compare the server time and time zone with the rule’s window, and check Study Time against arrival time for the affected studies.
- Fix
- Give each schedule an explicit time zone, use one window that crosses midnight, and keep servers and modalities on NTP.
RadFuzion
How RadFuzion helps
RadFuzion is built to show which layer failed and to hold studies while the problem is fixed. What it does for the failures above:
- Connection tests that go past C-ECHO. Testing a DICOM destination opens an association, runs C-ECHO and checks that the destination accepts storage for common SOP classes (CT, MR, CR, DX, ultrasound and secondary capture) without sending an image. Query sources are tested with a real C-FIND. Google Cloud destinations are tested for write permission. Each failed test says why, for example that the association was rejected and the calling AE may not be permitted.
- Health checks. DICOM destinations get a C-ECHO every minute and cloud destinations a check every 15 minutes by default.
- Holds and replay. When a destination is unreachable (timeout, refused connection, lost connection, DNS failure or aborted association), its studies are held and checked again about every five minutes. When the destination recovers, the held studies are sent. Temporary errors from a destination that is up, such as running out of resources, are retried in rounds: up to 20 by default, with the wait doubling to a one-hour cap. Errors that will not clear by themselves, such as an object a DICOM destination rejects, are marked failed with the reason. HTTP errors from a cloud store are retried in rounds first. A failed exam keeps its images until someone resends or deletes it.
- Large studies. For DICOM destinations the send timeout grows with the size of each batch. Mammography and whole-slide studies, and any file over about 100 MB, are sent one per batch with a timeout of at least 30 minutes, and you can set batch size, timeout and retries per modality.
- Transcode fallback. If a rule asks for a transfer syntax and the conversion fails, RadFuzion sends the original instead and records the fallback.
- Cloud store results. Answers from Google Cloud, Azure and other DICOMweb stores are read instance by instance, so a 200, 202 or 409 that lists failed instances is not counted as delivered for the whole request: only the instances the store didn’t take are marked failed and sent again. Instances the store already holds are counted as already there and reported as such. An Azure token is renewed before it expires and after a 401. A Google token is renewed on the next forward attempt. When a batch to Google Cloud or Azure fails, each instance is retried on its own so one bad object does not block the rest.
- Identity fixes by rule. Late images can keep the accession’s original Study UID, an accession can be generated when one is missing, and studies missing required tags can be held for a fix instead of forwarded.
- Test rules before they go live. Pick a recent exam or type tag values, and see every rule that matches and where the study would go, without sending anything.
- Schedules with a time zone. A rule’s time window can have its own time zone and can cross midnight.
- Alerts. Failed forwards and tag-validation holds can notify you by email, Slack, Google Chat or webhook, and an optional alert warns when the image-processing queue backs up.
One behavior to plan for: when images are held for their HL7 order and the order does not arrive before the timeout (15 minutes by default), the exam waits in Needs attention with its images kept, and staff send it without the order or delete it. An order that arrives late still releases it, and a setting can send these exams on without the order instead. Set the timeout to cover your real order lag.
For more, see the DICOM router page, how uploads to Google Cloud Healthcare work, and how RadFuzion keeps radiologists reading during PACS downtime.
See it in your workflow
Make failures visible before someone asks for a missing study.
RadFuzion tests destinations for real, holds studies while a destination is down, retries failed cloud batches one instance at a time, and lets you test rules without sending anything.