A technologist at a CT console pulls up the worklist, finds the patient, and starts the scan. Most of what the scanner knows about that patient and that exam got there by crossing from one standard to another. At many of the sites I’ve worked with, it left the RIS as an HL7 v2 message, was picked up by a separate worklist service, and was handed to the modality as a DICOM C-FIND response. (In IHE’s model the RIS serves the worklist itself; more on that below.) Three representations of the same order, two translations at broker sites, and a dozen fields that each have to land in exactly the right place.
When it works, nobody thinks about it. When it doesn’t, the symptom is usually downstream and misleading: a study that won’t reconcile, a prior that doesn’t prefetch, an accession number that doesn’t match the report. This post traces the crossing field by field so that when something arrives wrong at the scanner, you know which hop to look at.
The path#
flowchart TB
RIS["RIS / EHR
order placed
and scheduled"] -- "HL7, when a
separate broker
serves the worklist" --> IE["Interface engine"]
IE -- "HL7, possibly
transformed" --> MWL["Separate worklist SCP
(a broker,
a PACS, or a VNA)"]
RIS <-- "MWL: modality sends
C-FIND-RQ, worklist
items return as C-FIND-RSP
(IHE model: the RIS
is the worklist SCP)" --> MOD["Modality"]
MWL <-- "MWL: modality sends
C-FIND-RQ, items return
as C-FIND-RSP" --> MOD
MOD -- "C-STORE" --> PACS["PACS / archive"]
MOD -. "MPPS N-CREATE /
N-SET" .-> PPS["PPS Manager
(the PACS
or the RIS)"]
PPS -. "status forwarded to RIS,
image and report managers" .-> RIS
classDef sys fill:#1e293b,stroke:#64748b,color:#e2e8f0;
class RIS,IE,MWL,MOD,PACS,PPS sys;
The dotted lines are MPPS, Modality Performed Procedure Step: the modality reporting with N-CREATE and N-SET that an exam started, and then that it finished or was discontinued. In IHE’s model it goes to a Performed Procedure Step Manager, which can pass it on to the RIS, the image manager, and the report manager. At the sites I’ve worked with, that manager is usually the PACS, sometimes the RIS. What happens to the worklist entry next is local. In my experience, some worklists drop an entry when MPPS says completed, some when the RIS marks the exam done, some when the scheduled time has passed. Sites without a working path end up with worklists full of exams performed hours ago, and technologists who pick the wrong one.
Order, procedure, step#
One thing gets lost when people describe the worklist as a translated order. Under the patient and the visit, DICOM’s worklist model has three levels (PS3.4, Section K.6):
- The imaging service request, which is the order. Its identifier on the worklist is the accession number.
- One or more requested procedures under it, each with its own Requested Procedure ID and the Study Instance UID its images will carry.
- One or more scheduled procedure steps under each procedure, each with its own Scheduled Procedure Step ID, station, and start time.
The worklist returns one entry per scheduled step. For a plain chest x-ray there’s one of each, so the three IDs can look interchangeable. They aren’t, and an interface that treats them as the same works until the first exam that isn’t simple.
The field map#
Who fills each field matters as much as where it comes from. In IHE’s Scheduled Workflow, the RIS (IHE’s term is Department System Scheduler/Order Filler) isn’t a translator. It schedules the exam, generates the accession number, the requested procedure and step IDs, the Study Instance UID, the station, and the start time itself, and serves the worklist from its own database. The patient, visit and order data (identity, priority, reason for the exam, physicians, clinical alerts) come from the HL7 order; everything about scheduling is generated by the RIS (IHE RAD TF Vol 2x, Appendix B).
When a separate broker serves the worklist instead, it has to get those RIS-generated values from an HL7 message the RIS sends downstream. IHE defines no feed to a separate worklist broker. The nearest message is Procedure Scheduled (RAD-4), sent the first time a study is scheduled, and Procedure Update (RAD-13) for every change after that, both from the RIS to the image manager and the report manager; in my experience, brokers are often fed a copy. It carries the IDs, the UID, the modality and the start time, but not the station AE title, which a broker has to assign itself. Where each value travels depends on the HL7 version. In the v2.3.1 form (an ORM), the accession is in OBR-18, the Requested Procedure ID in OBR-19, the step ID in OBR-20, the modality in OBR-24, and the Study Instance UID in component 1 of ZDS-1, a Z-segment IHE defines for it. In the v2.5.1 form (an OMI), the same values move to the IPC segment, in the first component of each field: IPC-1 accession, IPC-2 Requested Procedure ID, IPC-3 Study Instance UID, IPC-4 step ID, IPC-5 modality, IPC-6 protocol code (RAD TF-2, Section 4.4.4.1.2). A broker that copies a whole field puts component separators into a DICOM value. That’s the closest thing IHE gives a broker to a standard layout, and a broker configured to it is easy to reason about. Many of the order feeds I’ve seen predate it or ignore it, and the broker’s configuration decides which field it reads.
So the table below says who fills each attribute first, then where it travels in HL7 under IHE (v2.3.1 and v2.5.1 where they differ), then what you’ll find in the field. I’ve kept to the attributes that cause trouble. “Top” means top level of the worklist response; “SPS” means inside the Scheduled Procedure Step Sequence (0040,0100).
| DICOM worklist attribute | Tag (where) | Who fills it, and where it travels in HL7 | What goes wrong |
|---|---|---|---|
| Patient’s Name | (0010,0010), top | From the order: PID-5 | Component order. HL7’s XPN is Family^Given^Middle^Suffix^Prefix; DICOM’s PN is Family^Given^Middle^Prefix^Suffix. The fourth and fifth components are swapped between the two standards, so an engine that copies the string through puts “JR” where DICOM expects “DR”. HL7’s degree component (XPN.6, “MD”) belongs in the DICOM suffix too, after the suffix (RAD TF-2x Appendix B, Note 10). Engines that reorder, truncate, or drop the caret separators do worse |
| Patient ID | (0010,0020), top | From the order: PID-3, component 1, from the repetition whose assigning authority (PID-3.4) and identifier type (PID-3.5) mark it as the MRN | PID-3 repeats: MRN, enterprise ID, insurance ID. “Take the first one” isn’t a rule; a registration change can reorder the repetitions, and every study lands under the wrong identifier |
| Issuer of Patient ID | (0010,0021), top | From the order: PID-3.4.1 (namespace ID). PID-3.4.2 and 3.4.3 and PID-3.5 go into the Issuer of Patient ID Qualifiers Sequence (0010,0024) as Universal Entity ID, its Type, and Identifier Type Code (PS3.3 Table 10-18) | HL7 carries the assigning authority as a structured value; DICOM splits it between Issuer of Patient ID and the Qualifiers Sequence. On the feeds I’ve seen, it’s often dropped entirely. Then the prior-matching problem begins |
| Patient’s Birth Date | (0010,0030), top | From the order: PID-7 | HL7 may carry a timestamp; DICOM DA wants YYYYMMDD. Engines that pass the time through produce an invalid DA, which some modalities reject |
| Patient’s Sex | (0010,0040), top | From the order: PID-8 | HL7 allows values DICOM doesn’t (U, A, N); DICOM’s enumerated values are M, F, O, or empty if unknown. IHE’s mapping is U to empty and A and N to O (RAD TF-2x Appendix B, Note 11). Engines without that mapping pass values the modality may reject |
| Placer / Filler Order Number | (0040,2016) / (0040,2017), top | Placer: ORC-2, from the order. Filler: ORC-3, assigned by the RIS. IHE’s Note 4 (RAD TF-2x Appendix B) has each carry the number and its issuer; current DICOM puts the issuer in Order Placer / Order Filler Identifier Sequence (0040,0026) / (0040,0027) (PS3.3 C.4.12) | Separate attributes from the accession in DICOM, even at sites that use the same value for both. Worth populating: they’re the link back to the order |
| Accession Number | (0008,0050), top | Generated by the RIS. Under IHE: OBR-18 (Placer Field 1) in v2.3.1, IPC-1 in v2.5.1. On many EHR-native feeds I’ve seen: ORC-3 / OBR-3 (filler order number) | The single most-argued field. On those feeds the accession sits in the filler order number and OBR-18 is empty. IHE itself expects IPC-1 to equal the filler order number in OBR-3 in most cases (RAD TF-2, 4.4.4.1.2.2.9); the trouble starts when OBR-18 or IPC-1 is empty and systems disagree which to read. DICOM gives it 16 characters, so a longer source number needs a mapping someone designed; silent truncation can give two orders the same accession. IHE also has the RIS fill the Issuer of Accession Number Sequence (0008,0051), which in v2.5.1 travels in IPC-1 components 2 to 4 under the Enterprise Identity Option (the v2.3.1 form has no place for it), and which matters for the same reason the issuer of patient ID does once two institutions share an archive. Which field the broker reads is configuration; if the RIS, the broker, and the PACS disagree, nothing reconciles |
| Requested Procedure ID | (0040,1001), top | Generated by the RIS. Under IHE: OBR-19 (Placer Field 2) in v2.3.1, IPC-2 in v2.5.1; otherwise the broker derives one | Not the procedure code: under IHE, OBR-44 carries the Requested Procedure Code, and OBR-4 components 1 to 3 keep the order’s code. This is an identifier for one requested procedure, the same in every OBR (v2.3.1) or IPC (v2.5.1) that belongs to it. Brokers that reuse the accession here are fine until one accession has two procedures |
| Scheduled Procedure Step ID | (0040,0009), SPS | Generated by the RIS when it schedules. Under IHE: OBR-20 (Filler Field 1) in v2.3.1, IPC-4 in v2.5.1; otherwise generated by the broker | It doesn’t exist when the order is placed, only once the exam is scheduled. The modality returns it in the MPPS, in the Scheduled Step Attributes Sequence (0040,0270); if the broker regenerates IDs on an order update, the receiver can’t match the MPPS to the scheduled step |
| Study Instance UID | (0020,000D), top | Generated by the RIS. Under IHE: ZDS-1, component 1 (IHE’s Z-segment for exactly this) in v2.3.1, IPC-3 in v2.5.1; otherwise generated by the broker | The UID the images will carry. If the RIS, the broker, and the modality can each mint one, you get the same exam under two UIDs and a reconciliation job that never ends. One system assigns it; everyone else copies it. That’s the scheduled case: an unscheduled exam has no worklist entry, so the modality mints the UID, and reconciliation has to attach the order to that UID rather than create another |
| Scheduled Station AE Title | (0040,0001), SPS | Generated by the RIS from its scheduling of rooms and equipment; not in the order. A broker assigns it from its own station table | Most modalities I’ve worked with filter on it. Wrong AE title here and the exam doesn’t appear on that scanner’s list. A room change that never reaches the worklist leaves the exam on the old scanner |
| Scheduled Procedure Step Start Date/Time | (0040,0002), (0040,0003), SPS | Generated by the RIS when it schedules. Under IHE: ORC-7, fourth component (v2.3.1) or TQ1-7 (v2.5.1); OBR-36 on some feeds | Timezone. HL7 timestamps may carry an offset; DICOM DA and TM don’t (only DT does, and Timezone Offset From UTC (0008,0201) is, in my experience, rarely populated on worklists). Converting to UTC isn’t wrong in itself; losing the local scheduling context is. A broker that converts to UTC and then serves the result as local time moves a 7 p.m. Central Standard Time exam to 01:00 the next day, onto tomorrow’s list. And OBR-7 is the observation date/time, not the scheduled one; engines confuse them |
| Modality | (0008,0060), SPS | Determined by the RIS from the order’s content, in DICOM defined terms. Under IHE: OBR-24 (Diagnostic Service Section ID) in v2.3.1, IPC-5 in v2.5.1, already as the DICOM defined term | On the non-IHE feeds I’ve seen it isn’t a direct copy; a mapping table decides that service section CT means modality CT. New procedure codes that miss the table fall through to a default or to nothing |
| Requested Procedure Description | (0032,1060), top | Generated by the RIS with the requested procedure. Under IHE: OBR-44, component 5 when sent (it’s optional), with laterality from OBR-15 (v2.3.1) or OBR-46 (v2.5.1) appended. OBR-4’s text is the order’s service name, not this | Free text. DICOM’s LO limits it to 64 characters, so a longer source description needs a deliberate shortening or mapping; silent truncation can drop clinically meaningful words |
| Scheduled Protocol Code Sequence | (0040,0008), SPS | Generated by the RIS. Under IHE: OBR-4, components 4 through 6 in v2.3.1, IPC-6 in v2.5.1 | Coded protocol for the step. A modality that supports protocol mapping can use it to preselect the acquisition protocol, but only if the codes agree on both ends. Most sites I’ve seen don’t populate it |
| Referring Physician’s Name | (0008,0090), top | From the order: PV1-8 (Referring Doctor) | Same component-order problem as patient name. The ordering provider (OBR-16 in IHE’s mapping; ORC-12 carries the same HL7 item) maps to a different attribute, Requesting Physician (0032,1032); sites I’ve seen that put it in both are guessing |
| Admission ID | (0038,0010), top | From the visit: PV1-19 (Visit Number), falling back to PID-18 (Account Number) when PV1-19 is empty. That’s IHE’s rule (RAD TF-2x Appendix B, Note 6); the issuer goes in Issuer of Admission ID Sequence (0038,0014) | Rarely needed on the worklist, frequently needed for reconciliation later; often not mapped |
One detail matters when you inspect a real response: not everything is at the top level. The station AE title, modality, step ID, start date and time, and protocol codes sit inside the Scheduled Procedure Step Sequence. A tool, or a person, looking for them at the top level reports them missing when they aren’t.
Three general patterns run through that table. Names and dates are format problems: the data is right, the encoding isn’t. Identifiers and the accession are selection problems: HL7 offers several candidates, and the site has to pick one and make every system agree. The generated IDs are ownership problems: exactly one system should create each one, and every other system should copy it. Format problems show up at the scanner. The other two show up weeks later, when someone tries to match a study to a report.
Where the engine changes things#
The interface engine in the middle is where most sites customize, and where much of the breakage lives, because the customizations are undocumented and were made by someone who left. The common ones:
- Repetition selection. Picking which PID-3 to forward. If the rule is “first one,” and a registration change reorders the repetitions, the MRN silently changes.
- Code translation. Mapping the RIS’s procedure codes to the broker’s, or to modality values. The table ages badly; every new procedure the RIS adds is a potential miss.
- Message filtering. Forwarding only
NW(new order) andCA(cancel) control codes, and droppingXO(change). Then a changed order (a new time, a different scanner) never reaches the worklist, and the technologist performs the old version. IHE’s own downstream update, Procedure Update (RAD-13, to the image manager and report manager), usesXOfor changes andCAorDCfor cancellations; a filter that dropsXOthere has the same effect. (Between the order placer and the RIS, IHE changes an order by cancel and new order instead.) - Field truncation and sanitization. Stripping characters the downstream system doesn’t like. Occasionally stripping the caret that separates name components.
A worklist isn’t built once. Every reschedule, room change, and cancellation has to arrive too, in time for the technologist to see it.
When a field arrives at the scanner wrong, compare it at each boundary: as the RIS sent it, as the engine passed it on, and as the worklist returned it. The hop where it changes is the system to look at. The engine is often that hop, but not always.
Testing it end to end#
The method that’s worked for me is a matrix rather than a checklist: order types across the rows (routine, stat, emergency with a temporary ID, outside or “foreign” order, order placed after the exam, exam moved to a different scanner after it was scheduled), modalities across the columns, and for each cell the fields that matter most for that combination. An emergency order on a trauma CT tests the temporary-patient-ID path and the later merge; a foreign order on an MR tests the issuer-of-patient-ID mapping; a routine CR tests the station AE title mapping, because in my experience CR rooms are where it’s most often wrong; a moved exam tests whether the reschedule reaches the worklist and the old scanner stops showing it.
For each cell:
- Place the order in the RIS (or inject the order message at the engine, with the RIS’s blessing).
- Capture the HL7 as it leaves the RIS and as it leaves the engine, and diff them. Every difference is a transformation someone configured.
- Query the worklist SCP with the same C-FIND the modality sends: its calling AE title, its matching keys (station AE title, modality, date range, whatever it uses), and its return keys. DCMTK’s
findscu -Wor pynetdicom with the worklist SOP class will do it. Compare each returned attribute to the table above, including the ones inside the Scheduled Procedure Step Sequence. - Perform the exam, or simulate MPPS with a tool. If the site uses MPPS, confirm the receiver recorded the start and the completion. Separately, confirm the entry leaves the worklist the way the site expects; that may or may not be MPPS’s doing.
- Store an image and check that the patient, accession, procedure, and study identifiers landed in the header as expected. Then check the PACS: did it file the study under the right patient and the right order? The header is what reconciliation works from; whether it worked is something you confirm in the receiving system.
Step 3 is the one most sites skip. They test from the scanner, which only shows whether the exam appears, not whether every attribute behind it is correct. A worklist entry with the right name and the wrong issuer of patient ID looks perfect on the console and causes trouble for years.
Sources#
- IHE Radiology Technical Framework, Volume 2x, Appendix B, HL7 Order Mapping to DICOM MWL (Rev. 23.0, August 2025), for which worklist values come from the order and which the Department System Scheduler generates, and Notes 4, 6, 10 and 11 for order numbers, Admission ID, name and sex mapping: https://www.ihe.net/uploadedFiles/Documents/Radiology/IHE_RAD_TF_Vol2x.pdf.
- IHE Radiology Technical Framework, Volume 2 (Rev. 23.0, August 2025), Scheduled Workflow transactions (RAD-2 Placer Order Management, RAD-3 Filler Order Management, RAD-4 Procedure Scheduled, RAD-5 Query Modality Worklist, RAD-6/7 MPPS, RAD-13 Procedure Update): https://www.ihe.net/uploadedFiles/Documents/Radiology/IHE_RAD_TF_Vol2.pdf. The HL7 field positions for the generated values are from RAD-4, Section 4.4.4.1.2: the v2.3.1 form (ORM; Tables 4.4-4, 4.4-6 and 4.4-7 for ORC, OBR and ZDS) and the v2.5.1 form (OMI; Tables 4.4-12 to 4.4-16, including the IPC segment). OBR-44 for the Requested Procedure Code and Description and OBR-15/OBR-46 for laterality are in the same OBR tables. RAD-4 versus RAD-13 is in Section 4.4.4.1.1.
- IHE Performed Procedure Step Manager actor: https://wiki.ihe.net/index.php/Performed_Procedure_Step_Manager.
- DICOM PS3.4, Annex K, Basic Worklist Management Service, Section K.6 for the worklist information model and which attributes sit in the Scheduled Procedure Step Sequence: https://dicom.nema.org/medical/dicom/current/output/chtml/part04/sect_K.6.html. Annex F (Procedure Step SOP Classes), Section F.7 for MPPS, including the Scheduled Step Attributes Sequence (0040,0270): https://dicom.nema.org/medical/dicom/current/output/chtml/part04/sect_F.7.2.html.
- DICOM PS3.3 for the Modality Worklist attributes and the Patient, Visit, and Study modules they draw from, including Section 10.15, Issuer of Patient ID Macro, Section C.4.12, Imaging Service Request Module (placer and filler order numbers and their issuer sequences), and Section C.7.1.1, Patient Module: https://dicom.nema.org/medical/dicom/current/output/chtml/part03/sect_C.4.12.html.
- HL7 v2.5.1, Chapter 2A, XPN and CX data types, for name component order and the assigning authority; DICOM PS3.5, Section 6.2, PN value representation, for DICOM’s name order.
- HL7 v2.5.1, Chapter 4 (Order Entry) for ORM/OMI and the ORC, OBR, TQ1 and IPC segments, Chapter 3 for PID and PV1.
- The failure patterns and the test matrix are from my own engagements and are stated as experience.
