Value Representation (VR)

DICOM supports two encoding paradigms that dictate whether the VR is present in the data stream:

• Explicit VR: The two-character VR code (e.g., 'PN', 'DA') is physically written into the element header, making the data set self-describing and easier to parse without prior knowledge of the IOD.
• Implicit VR: The VR is omitted from the data stream entirely. The parser must know the VR for every tag from the IOD definition or data dictionary to correctly interpret the length and value fields.

Explicit VR is mandatory for all DICOM file formats (Part 10) and is the dominant mode in modern network communication.

String VRs represent human-readable or coded text with specific character constraints:

• AE (Application Entity): Up to 16 characters identifying a DICOM node, with no leading/trailing spaces.
• CS (Code String): Uppercase alphanumeric with no spaces, used for controlled terminology values.
• LO (Long String): Up to 64 characters, often used for institution names or device identifiers.
• PN (Person Name): A complex type with caret-delimited components (Family^Given^Middle^Prefix^Suffix) supporting ideographic and phonetic representations.
• SH (Short String): Up to 16 characters.
• UI (Unique Identifier): A UID string with period-delimited numeric components, max 64 bytes.

Numeric VRs encode quantitative measurements and temporal data with strict formatting rules:

• DS (Decimal String): A string representing a fixed-point or floating-point number, used for measurements like slice thickness.
• IS (Integer String): A string of digits representing an integer value.
• DA (Date): Fixed 8-byte format YYYYMMDD, enabling direct string comparison for date range queries.
• TM (Time): Format HHMMSS.FFFFFF with variable fractional seconds precision.
• DT (Date Time): Concatenated date and time, e.g., 20240101120000.000000.

These VRs are critical for temporal query/retrieve operations where date range matching is performed lexicographically.

Binary VRs handle opaque data blobs and structured pixel information:

• OB (Other Byte): An unstructured byte stream with undefined content, used for pixel data in 8-bit images.
• OW (Other Word): A stream of 16-bit words, used for pixel data where each word must be byte-swapped according to the Transfer Syntax endianness.
• OF (Other Float): 32-bit IEEE floating-point words.
• OD (Other Double): 64-bit IEEE floating-point words.
• SQ (Sequence of Items): A container VR that nests zero or more data sets (Items) within a sequence, enabling hierarchical data structures like repeating acquisition parameters.
• UN (Unknown): Used when the VR is not known at encoding time; treated as a byte stream.

Robust DICOM parsers must handle VR discrepancies gracefully, as real-world data often contains non-conformant encodings:

• Private Tags (Odd Group Numbers): May have a VR of UN or be encoded with a VR that contradicts the private element's actual content.
• Retired VRs: Older VRs like 'SL' (Signed Long) or 'UL' (Unsigned Long) may appear in legacy data sets.
• Length Mismatches: The value length field may not match the VR's expected byte count, requiring parsers to rely on the explicit length rather than the VR-defined size.
• Zero-Length Values: Valid for all VRs, indicating a null or absent value, distinct from a zero-length string.

When DICOM data is serialized to JSON via DICOMweb (Part 18), VRs are represented as a vr property alongside the Value array:

• Each element becomes a JSON object with keys vr and Value (and optionally BulkDataURI for large binary payloads).
• PN values are expanded into an Alphabetic object with Given, Family, Middle, Prefix, and Suffix fields.
• SQ sequences become nested JSON arrays of data set objects.
• DA/TM/DT values remain as their string representations.

This JSON model eliminates the need for binary parsing but requires VR-aware serialization to preserve type fidelity.

String VRs handle textual data with specific character restrictions and length rules.

• AE (Application Entity): 16-byte max string for device names, no leading/trailing spaces
• CS (Code String): 16-byte max uppercase alphanumeric with underscores, no spaces
• SH (Short String): 16-char max for short identifiers like accession numbers
• LO (Long String): 64-char max for longer text like institution names
• ST (Short Text): 1024-char max for free-text notes
• LT (Long Text): 10240-char max for extended narrative content
• UT (Unlimited Text): Unlimited length for bulk textual data

Example: Patient ID (0010,0020) uses LO, allowing up to 64 characters for the identifier.

The PN Value Representation encodes human names with component groups separated by carets (^).

Structure: LastName^FirstName^MiddleName^Prefix^Suffix

• Supports up to 64 characters per component group
• Allows 5 component groups maximum
• Supports ideographic and phonetic representations via the = delimiter
• Handles multi-byte character sets for international names

Example: Doe^John^^Dr.^Jr. represents Dr. John Doe Jr.

Parsing Rule: Always split on ^ first, then handle = for alternate representations.

Temporal VRs enforce strict formatting to ensure unambiguous chronological parsing across systems.

• DA (Date): Fixed 8-byte YYYYMMDD format, e.g., 20240115
• TM (Time): Variable format HHMMSS.FFFFFF, e.g., 143025.500000
• DT (Date Time): Concatenated YYYYMMDDHHMMSS.FFFFFF&ZZZZ, e.g., 20240115143025-0500
• AS (Age String): 4-byte format nnnW, nnnM, nnnD, or nnnY for weeks, months, days, years

Critical Rule: DA values are always exactly 8 bytes. TM can be up to 16 bytes including fractional seconds. DT combines both with optional timezone offset.

Numeric VRs define precise binary encoding rules for quantitative measurements and identifiers.

• DS (Decimal String): 16-byte max string representing fixed-point or floating-point numbers, e.g., 123.456 or -0.75E-2
• IS (Integer String): 12-byte max string for signed integers, e.g., -1024
• US (Unsigned Short): 2-byte binary unsigned integer, range 0-65535
• SS (Signed Short): 2-byte binary signed integer, range -32768 to 32767
• UL (Unsigned Long): 4-byte binary unsigned integer, range 0 to 2^32-1
• SL (Signed Long): 4-byte binary signed integer
• FL (Floating Point Single): 4-byte IEEE 754 single precision
• FD (Floating Point Double): 8-byte IEEE 754 double precision

Parsing Note: DS and IS are string-encoded, while US, SS, UL, SL, FL, and FD are pure binary. Byte ordering depends on the negotiated Transfer Syntax.

These VRs handle opaque binary data, unique identifiers, and specialized medical data structures.

• OB (Other Byte): Byte stream for 8-bit pixel data or binary blobs, length defined by explicit Value Length
• OW (Other Word): 16-bit word stream, commonly used for 16-bit pixel data
• OF (Other Float): 32-bit IEEE 754 float stream
• OD (Other Double): 64-bit IEEE 754 double stream
• OL (Other Long): 32-bit integer stream
• UI (Unique Identifier): 64-byte max string for DICOM UIDs, e.g., 1.2.840.10008.5.1.4.1.1.2
• UN (Unknown): Opaque byte stream when VR is unknown, parsed based on negotiated Transfer Syntax
• AT (Attribute Tag): 4-byte ordered pair of 16-bit unsigned integers representing a DICOM Tag (Group,Element)
• SQ (Sequence of Items): Zero or more Data Set Items, enabling nested data structures

Key Distinction: OB is for 8-bit data, OW for 16-bit. UI must match the DICOM UID format with dot-separated numeric components.

DICOM data elements can be encoded with Explicit VR or Implicit VR, fundamentally changing how the VR is communicated.

Explicit VR:

• VR field is a 2-byte string (e.g., PN, DA, US)
• For OB, OD, OF, OL, OW, SQ, UC, UN, UR, UT: 2 reserved bytes follow VR, then 4-byte Value Length
• For all other VRs: 2-byte Value Length follows VR

Implicit VR:

• No VR field present in the data element
• VR must be determined from the DICOM Data Dictionary using the Tag
• Always uses 4-byte Value Length
• Used only in DIMSE network communication, never in DICOM files

Critical Rule: A DICOM Part 10 file must always use Explicit VR. Implicit VR is only valid in network associations where both SCU and SCP have agreed to it.