DICOM routing rules are conditional statements that decide which images leave a system and which destination receives them. The documentation reviewed here describes that filtering logic in useful detail across several products, and it describes rule priority in at least three of them. What it does not describe, in any of the retrieved passages, is a named loop-prevention mechanism. That absence matters for anyone designing a multi-hop imaging workflow, because a routing rule that forwards a copy of a study to a second system can, in principle, be matched again downstream. This article sets out what the sources do specify, attributes each vendor's claims to that vendor, and proposes a validation approach that treats loop behaviour as something to be tested rather than assumed.

What a Routing Rule Actually Contains

The VistA Imaging DICOM Gateway routing guide frames a rule as a comparison between an image property and a user-supplied value, with the comparison result deciding whether the rule applies [1]. A single when statement precedes all conditions, and if may optionally be used instead, and a rule can carry multiple conditions [1]. The destination named in a rule must be either a valid Imaging destination, such as a VistARad workstation, or a valid DICOM Storage SCP destination, such as an external PACS or film printer [1]. That two-destination-type distinction is a structural constraint on what a rule can point at, not a stylistic preference.

AdvaPACS documents a comparable but differently shaped rule object. Its configuration reference lists Name and Description, Source AE, Local AE, Bind AE, Destination AE, Priority, Level and Enable as the fields of a rule [2]. The same documentation defines the four device roles: Source AE is the remote AE that originally sent data, Local AE is the local AE that received it, Bind AE is the local AE that sends routed data onward, and Destination AE is the remote AE that receives it [3]. Separating the receiving local AE from the sending bind AE is the detail worth noticing, because it means the outbound association can originate from a different local identity than the inbound one.

Priority, Levels and the Sampling Trap

AdvaPACS states that rules are evaluated in priority order, with 100 the highest and 1 the lowest, and advises higher priorities for more specific rules and lower priorities for more general ones [3]. Its Level field offers Study, Series and Instance, where Study evaluates conditions at study level and routes the entire study on a match, Series evaluates per series, and Instance evaluates each individual instance [3]. A separate AdvaPACS page adds a sampling detail that implementation teams should read carefully: at Series level, conditions are evaluated only on the DICOM tag values of the first instance of each series, and a true result routes all instances of that series [2].

That sampling behaviour has a practical consequence. If a tag varies within a series, a Series-level rule keyed to that tag will not see the variation, because only the first instance is inspected [2]. VistA's guide records a related evolution in its own revision history: revision 5 notes a change to study-based rather than image-based evaluation logic, and revision 6 records that patch 53 introduced the ability to use the value 205,45, the Origin Index, in a routing rule [1]. The Origin Index reference is the closest any retrieved passage comes to a field that could be used to reason about where an image came from, but the guide does not describe it as a loop control.

Practical detail: a rule-inventory check before enabling a second hop

Before enabling any rule that forwards to a system which itself forwards onward, build a one-page inventory listing, for each rule, the source AE, the destination AE, the level, the priority and every DICOM attribute in its conditions. AdvaPACS documents exactly these fields as the rule's contents [2][3], and XNAT documents its rules as tag-plus-regex pairs [4], so the inventory can be assembled from configuration rather than guesswork. Mark any pair of rules where one system's destination appears as the other's source. For each marked pair, record whether the condition attributes are preserved unchanged by any transformation step in between, since Qvera documents tag-level mapping and transfer syntax changes [5]. Keep the inventory under version control alongside the rule export, and re-check it after any upgrade, because XNAT documents that a version change altered the priority position of its projectRules configuration [4].

Scripted and Regex-Based Routing

Not every product expresses rules as form fields. Qvera describes its DICOM Router as scripted, with condition nodes written in JavaScript and a built-in Code Wizard, and states that routing is therefore not limited to what is inside the study [5]. Qvera also states that the product implements DICOM network communication directly, acting as both SCP and SCU, and lists C-STORE, C-FIND, C-MOVE, C-GET and C-ECHO among supported operations, with AE title allow-listing available on an SCP listener [5]. Those are vendor claims about the product's own capabilities.

XNAT takes a third approach, defining custom project routing rules as a DICOM tag, a regular expression and an optional capturing-group index, formatted as (gggg,eeee):pattern:capturing-group-index [4]. XNAT's documentation gives a worked example using Study Description (0008,1030) with the pattern Project:\s*(\w+) to extract a project name [4]. It also records a version boundary: the dicom/projectRules configuration is the top priority in project routing in XNAT 1.8 and higher, whereas older installations treated it as fifth, and administrators who want the previous ordering can write custom rules that place the legacy settings first [4]. That is a documented behavioural change tied to a version, and it affects evaluation order rather than loop handling.

Where Loop Prevention Is and Is Not Documented

None of the retrieved passages names a loop-prevention feature. What the sources do describe is copy semantics and queue behaviour, both of which bear on how a cycle could arise. AdvaPACS states plainly that routing rules do not move the original uploaded data and instead send a copy of the matching study, series or instance to the configured Destination AE [3]. A copy that arrives at a second system is, from that system's perspective, newly received data, and the documentation reviewed does not describe a marker that would tell a downstream rule the copy has already travelled.

VistA's installation guide contains the most concrete operational warning in the set. When answering whether a computer will be part of a system where autorouting is active, the guide notes that a Yes answer creates queue entries for automated routing, and that if no Routing Processor is active at the site those queue entries will accumulate and never be processed or purged [7]. That is a documented failure mode of an unprocessed queue, not a loop, and the distinction should be kept. The routing guide also states that the age of an image is based on the transmission queue entry for that image rather than the date of the image itself [1], which means time-based conditions read queue age, not acquisition date.

A Validation Scenario Worth Running

Consider a suggested, hypothetical two-site arrangement: Site A runs a routing rule that sends copies of CT studies to Site B, and Site B runs a rule that sends copies of CT studies back to Site A. Nothing in the documentation reviewed states that either product would detect this. A defensible test is to configure the pair in a non-clinical environment, send a single identifiable test study, and observe whether the study count at each node stabilises or grows. The observation to record is the number of copies at each destination over a fixed interval, not a pass or fail judgement about the product.

The test design should also capture which fields each rule keys on. If Site A's rule matches on a tag that Site B preserves unchanged, the return rule has the same input it had the first time. Qvera states that its mapping nodes can read and write every DICOM value representation and change transfer syntax, and that UID transformation is deterministic, with the same input always producing the same output [5]. Deterministic UID handling is useful for keeping series linked after de-identification, but it also means a transformed identifier is reproducible, which is relevant when reasoning about whether a downstream rule will match again. That reasoning is inference from documented behaviour, not a documented loop control.

Ownership, Boundaries and Open Questions

Record ownership differs across the products in ways that shape who can change a rule. VistA's routing guide is written for imaging and radiology department staff responsible for setting up and maintaining a Routing Gateway [1], and the guide is a controlled US Government document that cannot be changed without consent of the development office [1]. XNAT places custom routing rules in the Site Administration UI, where an administrator enables site-wide custom routing and saves the configuration [4]. AdvaPACS exposes rules through a configuration interface with search, filter, clone, view, edit and create operations [2]. These are different governance surfaces, and the documentation does not rank them.

Several questions remain open on the evidence reviewed. No retrieved passage describes a hop counter, a visited-node list, a duplicate-suppression key or an explicit cycle check. MACH's DICOM Routing module description lists Verification, Storage, Storage Commitment and Modality Performed Procedure Step among supported service classes and describes TLS operation for incoming and outgoing connections [8], but says nothing about detecting repeated forwarding. aycan states that its auto-routing engine forwards incoming study images to destinations based on DICOM tag content down to the image level, and that rules can be scheduled for specific times or days [6], again without a documented cycle control. Treat loop behaviour as unverified until tested against the specific version in use.