.DCM File Format – Data elements

Tags

A tag is two unsigned 16-bit numbers, the group and the element, written (gggg,eeee) in hexadecimal. In a little-endian file the group is stored first, then the element, each least significant byte first: (0010,0010) is 10 00 10 00.

Element forms

The transfer syntax decides whether elements carry their VR (explicit VR) or not (implicit VR, where the reader looks the VR up in the dictionary). The two forms never mix in one data set.

• The value follows the length and always has an even number of bytes.
• A length of FFFFFFFF means undefined length: the value runs until a delimiter. It is allowed for SQ and UN, and for OB pixel data when the transfer syntax encapsulates it.
• An empty value has length 0.

Value representations

• Text VRs are padded to even length with one trailing space (20); UI is padded with one 00; OB with one 00 byte.
• Text uses the default character repertoire (ASCII) unless Specific Character Set (0008,0005) names another.

Value multiplicity

An element can hold several values. In text VRs they are separated by a backslash (5C), and the padding goes after the last value. Binary values of a fixed size are simply concatenated. LT, OB, OD, OF, OL, OV, OW, SQ, ST, UN, UR and UT always hold exactly one value.

Byte order

In little-endian transfer syntaxes, every binary number (tags, lengths, US, UL, SS, SL, FL, FD, OW, OF and so on) is stored least significant byte first. In the retired explicit VR big endian syntax they are stored most significant byte first, except OB and the group 0002 elements. Text values are byte strings either way.

Sequences and items

An SQ element holds items, each a nested data set. Items and delimiters use three special tags, always written in the implicit VR form (tag and 4-byte length, no VR), even in an explicit VR data set:

• A sequence of explicit length has a length covering all its items; an empty one has length 0. A sequence of undefined length ends with (FFFE,E0DD).
• Items inside the same sequence may mix explicit and undefined lengths.
• The data set inside an item follows the transfer syntax as usual, and may hold more sequences. An SQ element always has a value multiplicity of 1.
• Groups 0000, 0002 and 0006 are not allowed inside items.

08 00 15 11  53 51  00 00  FF FF FF FF      (0008,1115) SQ, undefined length
FE FF 00 E0  FF FF FF FF                    item, undefined length
   …data elements of the item…
FE FF 0D E0  00 00 00 00                    item delimitation
FE FF DD E0  00 00 00 00                    sequence delimitation

Private elements

Makers add their own elements in odd groups. Each maker reserves a block of 256 elements with a private creator element:

• (gggg,0010) to (gggg,00FF) are private creator elements, VR LO, value multiplicity 1, holding the maker’s identifying string.
• The creator at (gggg,00xx) owns the elements (gggg,xx00) to (gggg,xxFF): (gggg,0010) owns (gggg,1000-10FF), (gggg,0011) owns (gggg,1100-11FF), and so on.
• A creator string appears only once in a group.
• (gggg,0001-000F) and (gggg,0100-0FFF) are never used.

The meaning of private elements is up to each maker and is not part of the standard. In implicit VR files their VR is unknown to a reader without the maker’s documentation.

Common elements

The data dictionary (PS3.6) defines several thousand elements, too many to list here. These identify almost every object:

The elements that describe the image itself are on the pixel data page.