A PACS transfer syntax is not a compression setting; it is a set of encoding rules that lets two application entities negotiate byte ordering, data element structure, and compression before pixel data moves. DICOM PS3.5 defines the transfer syntax as an attribute that allows communicating systems to agree on common encoding techniques [1]. Compression enters the picture only when pixel data is sent in an encapsulated format, where the value representation OB is used and pixel cells are encoded according to a negotiated transfer syntax [1]. Evaluating a PACS therefore starts with the distinction between how data is represented in an association and how the underlying pixel stream is actually compressed. The two concepts interact, but they answer different questions: one concerns interoperability, the other concerns storage size, retrieval speed, and image fidelity. A rigorous evaluation treats them as separate decision points and then examines where they meet in retrieval and record management.

Native and Encapsulated Encoding Set the Evaluation Boundary

Pixel data conveyed in the Pixel Data element may be sent either in a native, uncompressed format or in an encapsulated, compressed format [3]. Native encoding carries restrictions that matter during evaluation. If Pixel Data is sent natively, the photometric interpretation must not be YBR_RCT, YBR_ICT, or YBR_PARTIAL_420, because those values are not encodable in uncompressed form [3]. Float Pixel Data and Double Float Pixel Data must also be native if encoded in a standard transfer syntax, unless future standard transfer syntaxes for compression of those elements are defined [3]. These constraints mean that a PACS evaluation cannot assume every image type is compressible under the same rules.

Encapsulated encoding changes the structural model. A frame may be entirely contained within a single fragment or may span multiple fragments to support buffering during compression or to avoid exceeding the maximum size of a fixed length fragment [3]. A recipient can detect fragmentation by comparing the number of fragments with the number of frames, and some performance optimizations may be available when fragmentation is absent, but an implementation that fails to support fragmentation does not conform to the standard [3]. For an evaluation team, this means compression support is not simply a yes or no capability. The ability to handle fragmented encapsulated pixel data is a conformance requirement that affects how gracefully a PACS handles large or streamed compressed objects.

Negotiation Defaults Differ Between DIMSE and Web Services

The default transfer syntax depends on the service boundary. For traditional DIMSE network communication, an application entity issuing an A-ASSOCIATE request must offer the DICOM Implicit VR Little Endian transfer syntax in at least one presentation context associated with each offered abstract syntax [2]. This requirement has a specific consequence: the same abstract syntax in two presentation contexts with two transfer syntaxes is not valid unless the default transfer syntax is also present in at least one presentation context based on that abstract syntax [2]. A PACS evaluation should verify that conformance statements reflect this negotiation structure rather than merely listing supported transfer syntaxes.

DICOMweb uses a different default. The default transfer syntax for DICOM objects contained in a payload is Explicit VR Little Endian Uncompressed, identified as 1.2.840.10008.1.2.1 [5]. This differs from the Implicit VR Little Endian default defined in PS3.5 [5]. Implicit VR Little Endian and Explicit VR Big Endian must not be used for DICOMweb payloads [5]. An origin server may support additional transfer syntaxes, and the transfer syntax can be one of the JPIP transfer syntaxes, in which case returned objects contain the URL of the JPIP provider for retrieving pixel data [5]. Evaluators comparing a PACS that serves both DIMSE and web clients must therefore check each interface against its own default and exception rules.

Checklist for Documenting Transfer Syntax and Compression Support

Record the conformance statement version and date, then list each storage SOP class with its declared transfer syntax set. Confirm that the DIMSE interface offers Implicit VR Little Endian in at least one presentation context per abstract syntax [2]. Confirm that the DICOMweb interface defaults to Explicit VR Little Endian and does not use Implicit VR Little Endian or Explicit VR Big Endian [5]. For encapsulated pixel data, verify that the system handles fragmented frames rather than assuming single-fragment encoding [3]. For lossy compression, document whether originals are retained, whether lossy copies use a separate series and new Series Instance UID, and what the system returns when a client cannot accept the lossy transfer syntax [4].

Lossy Compression Raises Retrieval and Record Ownership Questions

The DICOM standard explicitly places the clinical acceptability of lossy compression outside its scope. The context where lossy compression of medical images is clinically acceptable is beyond the DICOM Standard, and policies for selecting compression parameters such as compression ratio are also beyond the standard [3]. This is not a gap in the encoding specification; it is a deliberate boundary. A PACS evaluation that asks whether lossy compression is safe for primary diagnosis is asking a question the standard does not answer. The evaluation must instead examine what the system does with lossy objects and how it documents their provenance.

Retrieval behavior creates practical consequences. If a PACS or vendor neutral archive stores images only as lossy compressed and discards the original, then when requesting systems do not support lossy compressed objects for the particular SOP class, the PACS is required to decompress them to Implicit VR Little Endian [4]. That produces extra processing on the sending side and a larger, uncompressed image with any artifacts introduced by lossy compression [4]. If a PACS keeps both original and lossy versions, a study level retrieve returns both sets regardless of what is negotiated in the association, and no transfer syntax information is available in the query result [4]. One method is to generate lossy compressed objects in a separate series [4]. Lossy versions created afterward by a PACS or similar system should be stored in a separate series defined by a new unique Series Instance UID [4].

Conformance Statements Are the Primary Evaluation Artifact

A DICOM conformance statement is the document where a product declares which storage SOP classes and transfer syntax sets it supports. The conformance template instructs vendors to list all storage SOP classes in numerical order of SOP class UID and to indicate in the transfer syntax set column which of the defined transfer syntax sets are supported [7]. The template also notes that for each SOP class, multiple transfer syntax sets can be listed [7]. This structure lets an evaluator see not just whether JPEG lossless or JPEG lossy is supported, but which SOP classes carry that support.

The template defines example transfer syntax sets that can be modified to match a product implementation. The lossless compressed transfer syntax set includes JPEG Lossless, Non-Hierarchical, First-Order Prediction with UID 1.2.840.10008.1.2.4.70 and media type image/jpeg [7]. The uncompressed transfer syntax set includes Implicit VR Little Endian, Explicit VR Little Endian, and the retired Explicit VR Big Endian [7]. An evaluation should treat these as declarations, not proof of behavior. The conformance statement tells you what a vendor claims to support; it does not demonstrate that the implementation handles fragmented frames, converts to the required default, or preserves record identity under lossy re-encoding.

A Suggested Validation Scenario for Compression Evaluation

A useful hypothetical validation scenario would begin with a single CT study containing both native and encapsulated instances. The evaluation team would first confirm that the PACS conformance statement lists the expected transfer syntax sets for the CT Image Storage SOP class. Then the team would retrieve the study through the DIMSE interface and through the DICOMweb interface, recording which transfer syntax is actually returned in each case. The DIMSE response should reflect the negotiated presentation context, while the web response should default to Explicit VR Little Endian unless an exception applies [5].

The second part of the scenario would introduce a lossy compressed copy stored in a separate series with a new Series Instance UID [4]. The team would issue a study level retrieve and observe whether both the original and lossy series are returned, since the standard notes that a study level retrieve returns both sets regardless of what is negotiated [4]. The team would then request the lossy series from a client that does not support the lossy transfer syntax and verify that the PACS decompresses to the required default [4]. This scenario does not establish clinical acceptability; it establishes whether the system's declared transfer syntax behavior matches its observable retrieval behavior.