DICOM study instance UID reconciliation is best understood as a boundary problem rather than a single standardized operation. The DICOM Standard defines what a Study Instance UID is, how it must be encoded, and that semantic changes create new instances [8]. IHE Radiology assigns responsibility for generating that UID to the Order Filler [2]. PS3.17 then describes how inventory records for the same study may differ and what timestamp attributes can assist comparison, while explicitly warning that timestamp-based replacement is not always correct [7]. No source in this packet defines a universal reconciliation algorithm, a required merge transaction, or an automatic conflict-resolution rule. Implementers therefore combine stable identifier rules with local record-ownership and access-path decisions. The practical question is which system may change the identifier, which system may merge records, and what evidence supports each choice.

What the UID Itself Permits and Prohibits

A Study Instance UID is a globally unique identifier composed of an organization root and a suffix. The suffix must be unique within the scope of that root, and the same UID cannot identify both a study instance and a series instance [1]. The standard cautions implementers against building new UIDs by derivation, such as adding a suffix to a UID assigned by another implementation [1]. This constraint matters for reconciliation because a derived identifier can create false equivalence or collision when records from two systems are compared.

UIDs carry no semantics and must not be parsed to find a particular value or component [1]. Component definition for the suffix is implementation specific and may change as long as uniqueness is maintained [1]. Encoding rules require numeric components separated by periods, no leading zeroes except in a single-digit component, and a maximum of 64 total characters including any padding [1]. These rules define syntactic validity but do not tell a system how to decide that two records represent the same study.

The persistence rule is more consequential. Each instance is identified by a globally unique object identifier that persists across all exchanges, and changes to semantic content create a new instance with a new identifier [8]. For reconciliation, this means a corrected or altered study should not silently reuse an existing Study Instance UID if the semantic content changed. The standard does not, however, specify which metadata edits count as semantic changes in every deployment.

The information model adds a practical simplification: for acquisition modalities, a study is approximately equivalent to an ordered procedure, while in radiotherapy the study and series are less clearly related to real-world activities [8]. Reconciliation logic that assumes a one-to-one mapping between an order and a study may therefore be appropriate in radiology but less reliable in other domains. The standard states the model is sufficient for routine practice without claiming it resolves every domain-specific ambiguity [8].

Record Ownership and Generation Responsibility

IHE Radiology assigns a specific actor the job of creating the Study Instance UID. The Order Filler generates the Study Instance UID as a globally unique identifier for each Requested Procedure, and that identifier is used to identify all generated images and other DICOM objects related to the Requested Procedure [2]. This gives reconciliation a clear ownership anchor in scheduled workflow: the system that accepted the order is the documented source of the identifier.

The same framework distinguishes the Requested Procedure ID, which needs to be unique only within the scope of the Filler Order Number or Accession Number, from the Study Instance UID, which is globally unique [2]. A reconciliation design should not treat these identifiers as interchangeable. The local identifier may repeat across orders, while the UID must not. Confusion between the two can cause a merge routine to match records that belong to different studies.

Receiving systems also have documented responsibilities. The Image Manager or Image Archive should be prepared to receive, organize, and manage associated metadata for images from multiple studies sent on a single association, using their distinct Study Instance UIDs to associate images with the correct study [2]. This is a collation rule, not a merge rule. It tells the archive to preserve distinct UIDs when multiple studies arrive together, which is the opposite of collapsing them into one record.

The MADO supplement extends the ownership pattern to manifest documents. The Study Instance UID in the Imaging Study Manifest must use the same value as the referenced instances, and there is a one-to-one relationship between the manifest and the study it references [6]. A manifest that copies the study UID creates a second record type that must remain consistent with the original study. Reconciliation across a manifest and an archive therefore involves matching the same UID in different record classes rather than generating a new identifier.

What to Record Before Attempting a Merge

Before merging two records that share a Study Instance UID, capture the Item Inventory DateTime (0008,0404), Study Update DateTime (0008,041F), Content Date (0008,0023), and Content Time (0008,0033) from each record [7]. Compare the series sets first. If one record differs only by a later series, timestamp-based replacement may be reasonable; if the series sets differ entirely, do not auto-select by timestamp [7].

Also record whether either source uses the Removed from Operational Use attribute (0008,0405) with value Y, and whether the source retains series and instance records or deletes them [7]. This determines whether the consumer can distinguish deprecation from silent deletion. Keep the original Study Instance UID unchanged unless the semantic content of the study itself changed, in which case the standard's persistence rule points to a new instance with a new identifier [8].

Inventory Reconciliation as a Documented but Non-Algorithmic Task

PS3.17 addresses study record reconciliation directly, but as an informative operational consideration rather than a normative transaction. Within the tree of linked Inventory SOP Instances, a given study may be referenced multiple times among the Inventoried Studies Sequence Items, and each item is a complete record of the contents of the study as known by the creator of that item [7]. Differences in content are therefore expected, and the standard does not declare one record automatically authoritative.

Timestamp attributes may assist in establishing a timeline of changes to study content and metadata [7]. The document lists Content Date, Content Time, Item Inventory DateTime, Study Update DateTime, and Original Attributes Sequence as attributes that might be used [7]. A later record might reasonably be considered a more current replacement when it differs only by an additional series whose Series Date is after the prior Item Inventory DateTime [7].

The same section warns against overgeneralizing that rule. Two study records might have entirely different sets of series, and in that case simply choosing one record based on timestamp is probably not correct; the records would have to be further evaluated for the underlying reason for the difference and potentially merged in some way [7]. This is an explicit statement that reconciliation requires case-specific judgment, not a fixed timestamp preference.

Access paths add a second reconciliation dimension. A task for study record merge is reconciliation of access paths to stored SOP Instances of the study, which may present challenges if the study records link to different access methods, target folders, or container files [7]. In conflicting cases, it may be necessary to disregard study or series level access specifications and use only the access links to each SOP Instance as recorded in the instance level record [7]. This separates identifier matching from retrieval-path resolution.

Deletion, Deprecation, and the Limits of Consumer Knowledge

The deleted study example in PS3.17 shows how much reconciliation depends on source-system behavior. One system may retain study, series, and instance records with the Removed from Operational Use attribute set to Y. Another may mark the study deprecated and delete all series and instance data. A third may delete references but retain the study level database record without a deprecation flag. A fourth may delete all study information, leaving only an audit trail [7].

In the first two cases, the consumer application knows exactly what has happened and can determine whether to move the deprecated study data to a migration target repository [7]. That determination would be based on data retention policies and the technical approach of the target system [7]. In the cases where the source leaves no deprecation flag or no study record, the consumer has less information and must infer deletion from absence, which the document does not frame as a reliable reconciliation signal.

The DICOM data dictionary confirms the attribute-level identifiers involved. Study Instance UID has tag (0020,000D) and Series Instance UID has tag (0020,000E), both with UI value representation [5]. The dictionary also lists Study ID (0020,0010) as a separate SH attribute [5]. Reconciliation code that reads these attributes must distinguish the globally unique UID from the potentially local Study ID, a distinction that mirrors the IHE ownership discussion.

Web services expose the same identifiers through resource paths. The Retrieve Transaction defines Study Instances as /studies/{study}, Series Instances as /studies/{study}/series/{series}, and Instance as /studies/{study}/series/{series}/instances/{instance} [4]. These URI templates treat the study UID as a path segment, which means reconciliation logic operating over web APIs must handle UID values as opaque strings rather than parsed components, consistent with the encoding rule against parsing UIDs [1].

A Suggested Validation Scenario for Reconciliation Logic

A useful hypothetical validation scenario can test whether a reconciliation routine respects the documented boundaries without inventing a universal algorithm. Suppose an archive receives two inventory records for the same Study Instance UID. Record A contains two series and an Item Inventory DateTime of 10:00. Record B contains the same two series plus a third series whose Series Date is later than Record A's Item Inventory DateTime. Under the PS3.17 example, Record B might reasonably be considered a more current replacement [7].

The validation should then introduce a harder case. Record C contains an entirely different set of series under the same Study Instance UID. The routine must not simply choose the newer timestamp, because PS3.17 states that timestamp-based selection is probably not correct when the series sets differ entirely [7]. The expected behavior is to flag the records for further evaluation and potentially merge them in some way, with the merge policy supplied by the deployment rather than by the standard [7].

A third check should verify that the routine never derives a new Study Instance UID by adding a suffix to an existing one, since the standard cautions against building new UIDs by derivation from a UID assigned by another implementation [1]. If the routine proposes a corrected identifier, it should generate a new UID under its own organization root or reject the change. This check separates identifier repair from record matching.

Finally, the scenario should test access-path conflict. If Record A links to a study-level folder and Record B links to a different container, the routine should be able to disregard the study or series level access specifications and use only the instance-level access links [7]. A routine that passes these four checks demonstrates alignment with the documented constraints, but the scenario remains a suggested test, not a claim of independent validation or a required conformance procedure.