Content uploaded by David Clunie
Author content
All content in this area was uploaded by David Clunie on Oct 12, 2021
Content may be subject to copyright.
Content uploaded by David Clunie
Author content
All content in this area was uploaded by David Clunie on Jul 14, 2021
Content may be subject to copyright.
DICOM SR for communicating planar
annotations
An Imaging Data Commons (IDC) White Paper
David A. Clunie
2021/10/12
Overview
This white paper describes the technical details of best practices for coordinate-based
geometric description of planar annotations encoded in DICOM Structured Reports (SR) for
communicating user and machine created annotations about images, including quantitative
measurements and categorical statements about regions of interest and other image-related
geometrically defined entities. The intended audience is implementers familiar with the basic
principles of DICOM and DICOM Structured Reports. The SR 2D image-relative coordinates
content item (SCOORD) is described. Use of the SR 3D spatial coordinates content item
(SCOORD3D) to encode similar structures as points in space is also addressed. More complex
geometric constructs relevant to image annotation use cases and patterns beyond the primitive
content item types are described in terms of supporting PS3.16 templates. Particular attention is
devoted to the appropriate use of TID 1500 Measurement Report templates. Anonymous
content items without a concept name are also discussed, since they are frequently used for
coordinates and image references. Appropriate code selection is exemplified for various
scenarios. Issues of bounding boxes, interpolation and holes are also considered.
Table of Contents
Overview
Table of Contents
Introduction
Content Items, Relationships and Graphic Types
Value Type IMAGE
Relationship Type SELECTED FROM – Linking coordinates with images
Value Type SCOORD
Value Type SCOORD3D
Choice of Graphic Type
Single Points
Templates
Multiple Points
Templates
Single Line Segments – One or More Distance Measurements
Templates
Multiple Line Segments – Paths With or Without Distance Measurements
Templates
Interpolation
Closed Outlines and Geometric Objects
Templates
Bounding Boxes
Rectangles
Interpolation, Partial Pixels
Holes
Indication of Location, Direction and Magnitude – Arrow-like Things
Angles – With or Without a Shared Vertex
Anonymous Image and Coordinate Content Items
Tracking of Planar Annotations
Tracking Unique Identifier and Tracking Identifier
Observation UID
Referenced Content Items
Conclusions
References
Introduction
DICOM Structured Reports (SR) are the format of choice in the IDC platform for communicating
user and machine created annotations about images, including quantitative measurements and
categorical statements about subjects, cases, entire images, regions of interest (ROIs) [Wiki
ROI] and other image-related geometrically defined entities. Though ROIs may be encoded as
planar and volumetric bit planes in DICOM Segmentation objects (SEG) [DICOM SEG] and
referenced from SRs [DICOM SEG ref from TID 1410], this white paper describes coordinate-
based geometric description of planar annotations, primarily those encoded using the SR 2D
spatial coordinates content item (SCOORD) [SCOORD] using individual image-relative
coordinates. Use of the SR 3D spatial coordinates content item (SCOORD3D) [SCOORD3D] to
encode similar structures as points in space (but constrained to co-planar coordinates in a
single plane) will also be discussed.
The discussion of planar annotations will include not only geometrically-defined ROIs, but other
geometric constructs relevant to image annotation use cases. Some specific entities covered
include:
● geometrically-defined ROIs that nominally exactly correspond to the outline of an actual
entity (e.g., closed contour around a lesion) +/- “holes” within it
● geometrically-defined bounding regions that contain (but do not exactly define the outline
of) an entity (e.g., a “bounding box”)
● center(s) or approximate locations of one or more entities (on or inside a structure), each
defined by a point on an image (or in space)
● single line segments of finite size with meaningful endpoints (e.g., RECIST-style single
distance measurement)
● pairs of orthogonal line segments (e.g., WHO-style long and short axis bi-dimensional
measurements)
● triples of orthogonal line segments (e.g., for length, width and height measurements to
estimate volume), recognizing that by definition these are not coplanar
● multiple line segments defining a path (rather than an outline, i.e., not necessarily
closed) (for which there are coplanar and non-co-planar variants) (whether they are
interpolated or not)
● single line segments indicating a direction +/- significant magnitude +/- meaningful
endpoints, such as an arrow
● pair of line segments indicating two directions +/- meaningful endpoints (which may or
may not be co-located) (e.g., to measure an angle at a fixed point or an angle between
two lines without a shared vertex that is significant (such as a Cobb angle)
Specifically outside the scope of this discussion (and indeed SR encoding in general) are
semantically meaningless graphical objects that are intended to communicate only a specific
visual rendering for a human observer. A classic example is an arrow, which might
communicate a direction, possibly a magnitude, and possibly the location of its tip. From an SR-
encoded semantic perspective, each of these potential uses of an arrow would be distinguished
and encoded without concern for how it might be rendered, as opposed to a literal encoding of
the graphic elements of the line (such as thickness, arrow-head shape and size) that might be
stored in a DICOM Presentation State (PR) [DICOM PR] with no indication of its meaning (other
than by close proximity to a text box, perhaps).
Also outside the scope of this document are compound geometric representations, which might
be produced by combining multiple individual separately defined geometric regions. Though
there are efforts by the DICOM Radiotherapy (RT) working group (WG) to define so-called
“Conceptual Volumes” [DICOM Conceptual Volumes] using Boolean expressions and
references to SEG or RT Structure Set [DICOM RTSS] defined 3D-coordinate-based contours,
this mechanism has not yet been implemented, or extended to SR.
It will be assumed that the reader has a basic understanding of the principles of encoding
DICOM SR as a tree of content, each node of which consists of name-value pairs. The form of
the tree and the nodes are constrained by the various types defined for the nodes, such as
containers, codes, image references and spatial coordinates, the relationships that are possible
between them. These define the basic “infra-structure” of an “SR”, which is very flexible. To
achieve interoperability, it is necessary to further constrain the SR content to various
recognizable patterns. This is achieved through the definition of “templates” that define the
types, relationships and codes either for an entire SR object or for sub-parts of the SR-tree that
may be reusable for different use-cases. The remainder of this white paper will describe:
● the value types used to compose planar annotations
● how the existing standard templates may be used to encode planar annotations
● additional patterns for specific types of planar annotations that could be added to the
standard
This material derives from, corrects and extends the description of the SCOORD and related
content items and its use in [Clunie 2000], the discussion of geometric result encoding in [Clunie
2007], considers the practical implementation of clinical trial measurement templates described
in [Clunie 2011] and benefits significantly from a more recent discussion between the QIICR and
MINT Medical teams summarized in [Fischer 2018]. In particular, where possible the examples
have been designed to illustrate the use of TID 1500 Measurement Report [TID 1500] and its
subordinate TID 1410 Planar ROI Measurements and Qualitative Evaluations [TID1410] that is
used for planar regions of interest and TID 300 Measurement [TID 300], which in turn invokes
TID 320 Image or Spatial Coordinates [TID 320] for other planar measurements.
Content Items, Relationships and Graphic Types
In the following sections, a shorthand description of SR content items will be used, such as in
the following example
1
:
1
This is the shorthand output by the dicom3tools [dicom3tools] dcsrdump tool, and used for illustrative
purposes in [Clunie 2000]. It is not defined by the standard and is not a formal alternative representation
SCOORD: (111010,DCM,“Center”) = POINT {128.5,128.5}
SELECTED FROM: IMAGE: = (“NM Image”,“1.2.3.4”)
The Value Type, Relationship Type, Concept Name as a coded tuple, and a value appropriate
to the type
2
are shown. The level of nesting is expressed by the indentation.
A dump of the actual DICOM attribute encoding is as follows:
(0040,a040) Value Type “SCOORD”
(0040,a043) Concept Name Code Sequence
(fffe,e000) Item
(0008,0100) Code Value “111010”
(0008,0102) Coding Scheme Designator “DCM”
(0008,0104) Code Meaning “Center”
(fffe,e00d) Item Delimitation Item
(fffe,e0dd) Sequence Delimitation Item
(0040,a730) Content Sequence
(fffe,e000) Item
(0008,1199) Referenced SOP Sequence
(fffe,e000) Item
(0008,1150) Referenced SOP Class UID “1...”
(0008,1155) Referenced SOP Instance UID “1...”
(fffe,e00d) Item Delimitation Item
(fffe,e0dd) Sequence Delimitation Item
(0040,a010) Relationship Type “SELECTED FROM”
(0040,a040) Value Type “IMAGE”
(fffe,e00d) Item Delimitation Item
(fffe,e0dd) Sequence Delimitation Item
(0070,0022) Graphic Data “128.5,128.5”
(0070,0023) Graphic Type “POINT”
Value Type IMAGE
Fundamental to encoding image-relative (2D) coordinates is the ability to reference the image to
which they apply.
A value type of IMAGE (image object reference) contains a single reference to a DICOM image
object. The encoding of the value type is an extension of the COMPOSITE value type, which
contains a value that is encoded in a sequence, Referenced SOP Sequence (0008,1199), with a
single item that contains the Referenced SOP Class UID (0008,1150) and Referenced SOP
or encoding. Though the Code Value and Coding Scheme Designator are shown in this example as well
as the Code Meaning, in other examples they will be omitted for brevity.
2
In this shorthand example, as in others that follow, the UID for the SOP Class
(“1.2.840.10008.5.1.4.1.1.20” in this case) has been replaced with the more human readable
representation “NM Image” for clarity. In the SR object, it is encoded with the UID.
Instance UID (0008,1155) of the referenced image. SEG objects, which are encoded as bit-
planes in an image-like manner using the Pixel Data (7FE0,0010) attribute, are also referenced
using an IMAGE content item
3
.
There also may be:
● an optional Referenced Frame Number (0008,1160), which identifies one or more
frames as being referenced, if the reference is to a multi-frame image; if this attribute is
not included in a reference to a multi-frame image, then all the frames are referenced
● an optional Referenced Segment Number (0062,000B) if the reference is to a SEG
object, which identifies one or more segments as being referenced, if there is more than
one and the reference is not to all of them
● an optional reference to a Grayscale Softcopy Presentation State Storage object,
encoded in a nested Referenced SOP Sequence (0008,1199) containing exactly one
item
● an optional reference to a Real World Value Mapping object, encoded in a Referenced
Real World Value Mapping Instance Sequence (0008,114B) containing exactly one item
There may not be more than one item (since that would imply more than one value), and there
may not be zero items (since that would imply no value).
When an IMAGE value type content item is used, it may or may not have a concept name,
which specifies the purpose of the reference to the image
4
.
Relationship Type SELECTED FROM – Linking coordinates with
images
The SELECTED FROM relationship type specifies that the parent (source node) is a set of
coordinates selected from a child (target node) that is either another set of coordinates,
or an image or waveform.
This is encoded as shown in the following example:
SCOORD: (111010,DCM,“Center”) = POINT {128.5,128.5}
SELECTED FROM: IMAGE: = (NM Image,“1.2.3.4”)
There are several reasons that the coordinates are encoded in a separate node of the tree from
the image reference itself:
3
See, for example, reference from TID 1410 [DICOM SEG ref from TID 1410].
4
A concept name is not required however, and it is common for standard templates to allow the concept
name to be omitted (be a so-called “anonymous content item”), since the purpose of the reference to the
image is apparent from the surrounding context.
● Separating the coordinates from the image reference allows a single coordinate
reference to be applied to multiple images
5
. For example, one might specify a cylindrical
volume of interest through a block of contiguous parallel CT slices.
● To allow both spatial and temporal coordinates to be selected from the same image.
Value Type SCOORD
A value type of SCOORD (image-relative spatial coordinates) [SCOORD] contains a set of
spatial coordinates relative to the position within an image. The encoding of the value type is as
a set of points in Graphic Data (0070,0022) representing the x and y points of a geometric
object of the form specified in Graphic Type (0070,0023)
6
.
For example:
SCOORD: (111010,DCM,“Center”) = POINT {128.5,128.5}
An SCOORD content item will always have one or more children that are of IMAGE value type.
It is not possible to use an SCOORD to refer to an arbitrary point in space outside the context of
an image; there is a different value type, SCOORD3D, for that. Children are included by adding
a Content Sequence attribute, as with any other SR content item.
When an SCOORD content item is used, it may or may not have a concept name, which
specifies the purpose of the reference to the spatial coordinates
7
.
The choice of graphic types available for an SCOORD is:
● POINT, which specifies a single point
● MULTIPOINT, which specifies multiple, independent points
● POLYLINE, which specifies an open path or a closed polygon
8
defined by its vertices
● CIRCLE, which specifies a circle defined by its center and a point on the perimeter
● ELLIPSE, which specifies an ellipse defined by the endpoints of its major and minor
axes
No other values of Graphic Type (0070,0023) are permitted, since if an implementer could
extend these, they would not be interoperable.
5
An error in Supplement 23 [Sup 23] suggested that references to multiple images could be contained in
a single IMAGE value type. This was corrected by CP 216 [DICOM CP 216]. The mechanism intended for
applying the same coordinates to multiple images is to use multiple IMAGE value types.
6
This means of describing vector graphic objects is very similar to the Graphic Annotation Module of PR
objects.
7
A concept name is not required however, and it is common for standard templates to allow the concept
name to be omitted (be a so-called “anonymous content item”), since the purpose of the coordinates is
apparent from the surrounding context.
8
An important early clarification to whether or not POLYLINEs can be open and must be closed explicitly
by sending the same start and end point, or whether they are inherently defined to be closed, was added
in CP 233 [DICOM CP 233]. In short, they may be open or closed and are not assumed to be closed. The
description in [Clunie 2000] that predates CP 233 is incorrect, and is updated here.
The Graphic Data attribute consists of pairs of points, the first of the pair being the X or column
location, and the second being the Y or row location. These are encoded as floating point
values
9
with sub-pixel resolution defined relative to an image or frame of an image, though they
are not constrained to be within the image area. This means that:
● the top left hand corner of the top left hand pixel is 0.0,0.0
● the center of the top left hand pixel is 0.5,0.5
● the bottom right hand corner of the top left hand pixel is 1.0,1.0
The coordinates are not normalized to a particular range like 0.0 to 1.0, but rather are specified
in units of actual pixels in the image. So, for example, given an image with 512 columns and
256 rows:
● the top left hand corner of the bottom right hand pixel is 511.0,255.0
● the center of the bottom right hand pixel is 511.5,255.5
● the bottom right hand corner of the bottom right hand pixel is 512.0,256.0
This is illustrated in Figure 1.
Figure 1. Spatial Coordinates With Sub-Pixel Resolution
Since the coordinates are image specific, rather than normalized, they have no meaning unless
applied to a specific image. There can be no change in geometry of the referenced image,
otherwise the coordinates would be invalidated. Image-relative coordinates are equally
applicable to 3D images from cross-sectional modalities (like CT, MR and PET), and 2D images
(such as from photographs, projection radiography and planar NM). However, given an image-
relative set of coordinates, if the referenced image does contain patient-relative spatial location
9
These are IEEE 754:1985 32 bit binary floating point values with a DICOM VR of Float (FL), and not
decimal strings (DS), as are some other values in DICOM SR, such as measurements in NUM content
items.
information (as is the case cross-sectional modality images)
10
, the two-dimensional image-
relative coordinates can be mapped into patient-relative three-dimensional coordinates, even
though they are not directly encoded
11
.
Value Type SCOORD3D
A value type of SCOORD3D (3D spatial coordinates) [SCOORD3D] contains a set of 3D spatial
coordinates relative to a defined 3D Cartesian coordinate system. The encoding of the value
type is as a set of points in Graphic Data (0070,0022) representing the x, y and z points of a
geometric object of the form specified in Graphic Type (0070,0023).
The coordinate system is identified by a Frame of Reference encoded as a UID in the
Referenced Frame of Reference UID (3006,0024). Typically, this is a patient-relative coordinate
system, as used in DICOM cross-sectional images such as CT, MR and PET, but it is not
required to be. Patient-independent coordinate systems are defined in DICOM for other
applications, including volumetric ultrasound and whole slide microscopy.
For example:
SCOORD3D: (111010,DCM,"Center") = POINT {0,-8.04,-567.5} (FoR 1.2.3.4)
An SCOORD3D content item does not have any IMAGE children, since it refers to arbitrary
points in space, which may or may not be represented on any actual image.
When an SCOORD3D content item is used, it may or may not have a concept name, which
specifies the purpose of the reference to the spatial coordinates
12
.
10
For such cross-sectional modality images, DICOM defines a patient-relative Cartesian coordinate
space, with respect to an arbitrary origin, which is consistent within a defined frame of reference,
identified by a UID, and which typically has the scope of a single acquisition (series or study). For
example, a patient will be positioned on the table of the scanner, a reference point identified visually by
the operator, and then images acquired, and encoded with the common frame of reference UID. The
necessary frame of reference, orientation, origin position and spacing information is not only present in
the image attributes, but can also be encoded within the SR object by using the appropriate (optional)
Image Library entries [Cross-Sectional Image Library], to avoid the need for the recipient to retrieve the
image metadata. When it is necessary to integrate 2D outlines across contiguous parallel slices, to
compute a volume for example, one can derive the reconstruction interval between the centers of slices
from the TLHC voxel positions of the referenced images projected onto the normal to the row and column
orientation. For such purposes the reconstruction interval should always be computed, since though
slices may be parallel and contiguous they may not be equally spaced, and furthermore, the interval may
be different from the nominal slice thickness (i.e. there may be a gap or overlap).
11
Without indirection “through an image”, no three-dimensional points may be specified using the
SCOORD content item; instead, the SCOORD3D content item supports the direct encoding of three-
dimensional points. RTSS also uses image-independent three-dimensional locations.
12
A concept name is not required however, and it is common for standard templates to allow the concept
name to be omitted (be a so-called “anonymous content item”), since the purpose of the coordinates is
apparent from the surrounding context.
The choice of graphic types available for an SCOORD3D is:
● POINT, which specifies a single point
● MULTIPOINT, which specifies multiple, independent points, which need not be coplanar
● POLYLINE, which specifies an open path or a closed polygon defined by its vertices,
which need not be coplanar
● POLYGON, which specifies a closed polygon defined by its vertices, which shall be
coplanar
13
● ELLIPSE, which specifies an ellipse defined by the endpoints of its major and minor
axes
● ELLIPSOID, which specifies an ellipsoid defined by the endpoints of its three axes
No other values of Graphic Type (0070,0023) are permitted, since if an implementer could
extend these, they would not be interoperable.
The Graphic Data attribute consists of pairs of points, the first of the pair being the X or column
location, and the second being the Y or row location. These are encoded as floating point
values.
Choice of Graphic Type
The choice of graphic type to use depends on the application.
Single Points
The POINT graphic type may be used to indicate a single location of significance, such as the
center of the lesion. Both 2D and 3D points are supported
14
.
For example
15
:
13
Even though the SCOORD3D POLYGON graphic type is defined to be closed, it is still required that
the first and last end-points be the same and that they both be sent. This makes it consistent with a
POLYLINE representation of the same object, whether it be 2D (in an SCOORD) or 3D (in an
SCOORD3D).
14
In some cases only 3D coordinates are possible, e.g., when no image exists or the point was defined in
the absence of or separate from an actually encoded image, e.g., on a 3D visualization. In other settings,
only an image exists, and there is no defined 3D coordinate system, e.g., an annotation of a photograph
or projection radiograph. In other, very common cases, the choice is a matter of style, personal
preference and convenience. E.g., for cross-sectional modalities such as CT, MR and PET, typically 2D
image-relative (slice-relative) coordinates are captured by image oriented systems that do not perform 3D
volume extraction and visualization, and 3D coordinates by those that do. Note that in all cases of image-
relative coordinates defined on images that have an established 3D frame of reference, the 3D
coordinates can easily be computed by recipients that require them. The converse is not always the case,
since the 3D coordinates may not map to an actual 2D image location, nor a set of coordinates to a
renderable structure that is coplanar with a single 2D image. Historically, 2D coordinates have long been
used in DICOM SR and Presentation States, and 3D coordinates are less frequently encountered and
supported.
15
In these and the following examples of SCOORD, the required child SELECTED FROM IMAGE content
item is elided for clarity.
SCOORD: (111010,DCM,"Center") = POINT {128.5,128.5}
SCOORD3D: (111010,DCM,"Center") = POINT {0,-8.04,-567.5} (FoR 1.2.3.4)
Templates
This pattern, with the concept name of (111010, DCM, “Center”), was first used in the standard
for the CAD templates
16
for the communication of the location of image findings such as
microcalcifications. It may be reused with the TID 1500 Measurement Report [TID 1500] by
including within TID 1501 Measurement and Qualitative Evaluation Group [TID 1501] an
SCOORD or SCOORD3D content item even in the absence of a measurement or a defined
ROI
17
.
A similar pattern is supported by TID 1410 Planar ROI Measurements and Qualitative
Evaluations [TID1410] as invoked by TID 1500 Measurement Report [TID 1500]; specifically:
● The concept name for the SCOORD is required to be (111030, DCM, "Image Region"),
the same as for any other geometrically defined TID 1410 ROI.
● The content item description specifically calls attention to the fact that an infinitely small
area, such as the center of a lesion, may be described using a Graphic Type of POINT
18
.
● The (130400, DCM, "Geometric purpose of region") with a value of (111010, DCM,
“Center”) can optionally be used to signal whether the infinitely small region defined by
the point is the center of an entity, as opposed to some an arbitrary point somewhere
within whatever the region nominally represents.
For example:
SCOORD: (111030,DCM,"Image Region") = POINT {128.5,128.5}
Points may have measurements, for example derived from the signal intensity at a point on the
image or in space
19
, using TID 300 [TID 300] as invoked by TID 1501 [TID 1501] :
NUM: (126312,DCM,"Ktrans") = 0.0185 (/min,UCUM,"/min")
INFERRED FROM: SCOORD: = POINT {128.5,128.5}
16
See TID 4021 Mammography CAD Geometry [TID 4021], TID 4107 Chest CAD Geometry [TID 4107]
and TID 4129 Colon CAD Geometry [TID 4129]. The first two only describe 2D points, whereas the colon
CAD template allows either SCOORD or SCOORD3D single points.
17
CP 1999 [DICOM CP 1999] extended TID 1501 to allow image, waveform and coordinate references
despite the absence of a measurement.
18
Though this was permitted from the first introduction of ROI templates by CP 1112 [DICOM CP 1112]
and TID 1500 templates by CP 1386 [DICOM CP 1386], specific attention to the use of POINT for this
purpose was added in [DICOM CP 1789].
19
It is understood that the accuracy of signal intensity values may be subject to the manner in which it
was computed, and that this may be subject to many factors, including but not limited to the mechanism
of interpolation used, if any. The standard templates do not require that such measurement method
information be provided, but do allow for such modifiers.
Or TID 1410 [TID 1410]:
CONTAINER: (125007,DCM,"Measurement Group")
CONTAINS: SCOORD: (111030,DCM,"Image Region") = POINT {128.5,128.5}
NUM: (126312,DCM,"Ktrans") = 0.0185 (/min,UCUM,"/min")
Multiple Points
The MULTIPOINT graphic type may be used to indicate the locations of the centers of entities of
a similar type. Both 2D and 3D points are supported.
For example:
SCOORD: (111010,DCM,"Center") = MULTIPOINT {128.5,128.5,137.5,201.5}
Templates
One use case for multiple points is to locate entities that are grouped or clustered together, such
as microcalcifications on a mammogram. There is no requirement that the points be grouped,
however, and the same mechanism could be used to locate objects spread throughout an
image, such as nuclei on a histopathology image
20
.
Currently, there are no standard templates that make use of MULTIPOINT.
A more common means of communicating a cluster of entities, such as a cluster of
microcalcifications on a mammogram, is to define a closed region surrounding them using a
POLYLINE, CIRCLE or ELLIPSE
21
.
Single Line Segments – One or More Distance Measurements
The POLYLINE graphic type may be used to indicate the locations of the end points of a single
line segment. Both 2D and 3D points are supported.
20
There is a practical limit on the number of points that may be encoded within a single SCOORD or
SCOORD3D content item, due to the length of the Graphic Data array being of FL (IEEE 32-bit float)
value representation, with a maximum length of 216-4 bytes imposed by the 16 bit value length field, which
equates to 8191 2D points or 2730 3D points. WG 26 is considering this issue, and will likely come up
with an alternative representation for very large numbers of points or objects such as the nuclei use case
described.
21
Such is the approach used in mammography CAD, with TID 4021 [TID 4021] using (111041, DCM,
"Outline") as the concept name to distinguish the outlined area of a cluster from its center, and (111059,
DCM, "Single Image Finding") = (129769006, SCT, “Calcification Cluster”) in templates such as TID 4006
[TID 4006] to describe them.
For example:
SCOORD: (121055,DCM,"Path") = POLYLINE {17.8,27.1,13.8,34.0}
Templates
The intent can be communicated either with a specific concept name to indicate the purpose, or
deduced from context, such as if the coordinate content item is specified as the source of a
linear distance measurement
22
.
For example, a simple length measurement could be encoded as:
NUM: (410668003,SCT,"Length") = 6.4 (mm,UCUM,"mm")
INFERRED FROM: SCOORD: = POLYLINE {17.8,27.1,13.8,34.0}
Alternatively, more specific concept names for the measurement are available to use as
appropriate for single line segments, including a subset of those listed in CID 7470 Linear
Measurements [CID 7470]
23
, many of which have significance if used together for associated
measurements, e.g.:
● (81827009, SCT, “Diameter”)
● (103339001, SCT, “Long axis”) and (103340004, SCT, “Short axis”)
● (410668003, SCT, "Length"), (121207, DCM, “Height”)
24
and (103355008, SCT, “Width”)
The general measurement template TID 300 [TID300] is used for many different applications.
Image measurement reports encoded using [TID 1500], supports this pattern using TID 300
[TID300]. The pattern may be repeated for multiple associated measurements, as well as linear
distance measurements associated with an ROI
25
. An example of using this pattern to describe
22
It may be the case that a generic line drawing or simple linear distance measurement tool in an
application has no means of obtaining from the user a concept that describes the meaning, purpose or
intent of the drawing the line or making the measurement. A more sophisticated application might provide
a drop-down list of choices of coded concepts to use for such measurements, and may or may not be
configurable to extend the codes available. A computational application that generates such a
measurement automatically may have highly specific semantic information to convey.
23
TID 1500 [TID 1500] invokes TID 1501 [TID 1501] (which in turn invokes TID 300 [TID 300]) with a
$Measurement parameter that specifies the baseline context group CID 218 Quantitative Image Features
[CID 218], which in turn ultimately includes CID 7470 [CID 7470].
24
It is not clear why SNOMED CT lacks a concept for “height” as a general adjectival qualifier value,
similar to “width”, or a property of measurement qualifier value, similar to “length” or “depth”, or why
“width” is in a different class hierarchy from “length” or “depth” (or why there is not a different “width”
concept that is a property of measurement qualifier value.
25
A common use case in tumor response assessment is to make repeated measurements of lesion size
over time, measuring the maximum transverse (axial) diameter (unidimensional, RECIST-style), with or
without an additional measurement perpendicular to it on the same image (bidimensional, WHO-style).
Increasingly such applications are partially or fully automated, with a planar or volumetric ROI being
segmented, characterized, then linear distances derived from the segmentation automatically. The
segmented ROI, defined by contours or bitplanes, measurements of its volume, area and other
long axis (RECIST) and short axis (WHO bi-dimensional) is provided in PS3.17 [PS3.17
RRR.5], an extract from which follows:
NUM: (103339001,SCT,"Long Axis") = 9.21 (mm,UCUM,"mm")
CODE: (370129005,SCT,"Measurement Method") = (126081,DCM,"RECIST 1.1")
INFERRED FROM: SCOORD: (121112,DCM,"Source of Measurement") = POLYLINE
{x1,y1,x2,y2}
SELECTED FROM: IMAGE: = (CT Image,“1.2.3.4”)
NUM: (103340004,SCT,"Short Axis") = 6.8 (mm,UCUM,"mm")
CODE: (370129005,SCT,"Measurement Method") = (112029,DCM,"WHO")
INFERRED FROM: SCOORD: (121112,DCM,"Source of Measurement") = POLYLINE
{x3,y3,x4,y4}
SELECTED FROM: IMAGE: = (CT Image,“1.2.3.4”)
The same approach can be extended to a third dimension. The ultrasound family of
measurement templates go so far as to define a specific template for volume estimates from
three orthogonal linear distance measurements, TID 5016 LWH Volume Group [TID 5016], but
its highly specific structure is not compatible with the more general TID 1500 [TID 1500]
approach. Using the latter, the individual distance measurements and their coordinates can be
encoded in the normal manner using TID 300 [TID300] invoked by TID 1501 [TID 1501]. The
computed volume measurement (without its own coordinates) can be encoded similarly, though
the explicit derivation of the computed volume to its source linear measurements is not made
explicit
26
; that may be apparent from the concept name of the volume measurement, or use of a
modifier describing the method. For example:
NUM: (410668003,SCT,"Length") = 9.21 (mm,UCUM,"mm")
INFERRED FROM: SCOORD: = POLYLINE {x1,y1,x2,y2}
NUM: (103355008,SCT,"Width") = 6.8 (mm,UCUM,"mm")
INFERRED FROM: = POLYLINE {x3,y3,x4,y4}
NUM: (121207,DCM,"Height") = 4.2 (mm,UCUM,"mm")
INFERRED FROM: = POLYLINE {x5,y5,x6,y6}
NUM: (121221,DCM,"Volume of ellipsoid") = 137.7 (mm3,UCUM,"mm3")
or:
NUM: (410668003,SCT,"Length") = 9.21 (mm,UCUM,"mm")
INFERRED FROM: SCOORD: = POLYLINE {x1,y1,x2,y2}
NUM: (103355008,SCT,"Width") = 6.8 (mm,UCUM,"mm")
quantitative features, as well as the derived distance measurements, may all be recorded using [TID
1500], as exemplified by PS3.17 RRR.5 [PS3.17 RRR.5].
26
Though there is a (126011, DCM, "Derived Imaging Measurements") CONTAINER in TID 1500 [TID
1500] that uses TID 1420 Measurements Derived From Multiple ROI Measurements [TID 1420], which
allows references to regions of interest encoded elsewhere in the content tree as the source of derived
measurements, it does not currently support reference to individual measurements or those within TID
1501 [TID 1501]. It also results in a directed acyclic graph rather than a simple tree, which may be
undesired complexity.
INFERRED FROM: = POLYLINE {x3,y3,x4,y4}
NUM: (121207,DCM,"Height") = 4.2 (mm,UCUM,"mm")
INFERRED FROM: = POLYLINE {x5,y5,x6,y6}
NUM: (118565006,SCT,"Volume") = 137.7 (mm3,UCUM,"mm3")
CODE: (370129005,SCT,"Measurement Method") = (126029,DCM,"LWH method for
volume of ellipsoid")
Multiple Line Segments – Paths With or Without Distance Measurements
The POLYLINE graphic type may be used to indicate the locations of the intermediate and end
points of multiple line segments describing a path on an image (2D) or in space (3D).
For example:
SCOORD: (121055,DCM,"Path") = POLYLINE {17.8,27.1,13.8,34.0,15.2,24.3}
The SCOORD3D POLYLINE points are not required to be coplanar, nor need they exist on an
actual image.
Multiple line segments may be used for applications that involve measurements, or for those
that only require describing a path, such as the centerline or edges of vessels, organs, cavities
or other structures
27
.
Templates
As with single line segments, the intent can be communicated either with a specific concept
name to indicate the purpose, or deduced from context, such as if the coordinate content item is
specified as the source of a linear distance measurement.
For example, a simple length measurement could be encoded as:
NUM: (410668003,SCT,"Length") = 6.4 (mm,UCUM,"mm")
INFERRED FROM: SCOORD: = POLYLINE {17.8,27.1,13.8,34.0,15.2,24.3}
Alternatively, more specific concept names for the measurement are available to use as
appropriate for multiple line segments, including a subset of those listed in CID 7470 Linear
Measurements [CID 7470], e.g.:
● (410668003, SCT, "Length")
● (121211, DCM, “Path length”)
27
For example, TID 3214 Analyzed Segment [TID 3214], used for quantitative arterial analysis, uses
SCOORD concept names such as (122507, DCM, "Left Contour") and (122508, DCM, "Right Contour") to
define the borders of arteries, without having an associated measurement. TID 4008 Mammography CAD
Breast Geometry [TID 4008] uses (111007, DCM, "Breast Outline Including Pectoral Muscle Tissue") and
(111045, DCM, "Pectoral Muscle Outline").
● (121206, DCM, “Distance”)
Reports encoded using [TID 1500] supports this pattern by allowing the use of TID 300 [TID300]
when measurements are present. When measurements or a defined ROI are not present, TID
1501 Measurement and Qualitative Evaluation Group [TID 1501] can include a multi-segment
POLYLINE SCOORD or SCOORD3D content item.
Interpolation
The standard does not currently specify the use of any interpolation between the encoded
points of the line, which are implied to be joined by straight line segments
28
. In other words, a
jagged rather than smoothed curve is described. Conversely, the use of interpolation is not
expressly prohibited
29
, whether it be for the rendering of the polyline for display, or for
measurements derived from it. In the absence of a specific attribute of the SCOORD defining
the use of interpolation, one might consider using the concept name of the concept to signal
interpolation of some type. E.g., something like:
SCOORD: (,,"Interpolated Path") = POLYLINE {17.8,27.1,13.8,34.0,15.2,24.3}
However, such an approach might not generalize well to other uses of POLYLINE, such as
closed polylines for ROI measurements, particularly when they have template-defined concept
names such as (111030, DCM, ”Image Region”). Theoretically, most SR IODs allow any type
(including SCOORD and SCOORD3D) to have children with a HAS CONCEPT MOD
relationship, so a post-coordinated approach might also work, and possibly allow for a variety of
types of interpolation:
SCOORD: (121055,DCM,"Path") = POLYLINE {17.8,27.1,13.8,34.0,15.2,24.3}
HAS CONCEPT MOD: CODE: (,,"Interpolation method") = (,,"Cubic spline")
Another interpolation issue to consider is whether it might apply to the coordinates themselves,
or instead to the measurement. I.e., would the interpolation technique modifier be applicable
only to the measurement calculation and not to the line defined by the user and rendered on the
display, which might remain as a series of straight line segments, and hence might be encoded
as:
NUM: (410668003,SCT,"Length") = 6.4 (mm,UCUM,"mm")
28
This is different from the approach used in DICOM Presentation States, for which two distinct Graphic
Types are defined, POLYLINE and INTERPOLATED. The latter is defined as a “list of end points between
which some form of implementation-dependent curved lines are to be drawn”. See [DICOM PS3.3
C.10.5.1.2].
29
Though it probably should have been. The discussion of interpolation at the time considered that
different implementations of interpolation would produce different results (visually and semantically) and
no consensus could be reached on a suitable inventory of interpolation methods. It was also expected
that when it mattered, a higher sampling rate for the points, matching or exceeding the pixel resolution,
might mitigate this issue to some extent.
HAS CONCEPT MOD: CODE: (,,"Interpolation method") = (,,"Cubic spline")
INFERRED FROM: SCOORD: = POLYLINE {17.8,27.1,13.8,34.0,15.2,24.3}
Closed Outlines and Geometric Objects
The SCOORD POLYLINE, CIRCLE, ELLIPSE and the SCOORD3D POLYLINE, POLYGON,
ELLIPSE and ELLIPSOID graphic types may be used to outline objects on an image (2D) or in
space (3D).
Applications include indicating the source of size measurements such as circumferential linear
distance, area or volume, or other quantitative measures derived from contained signal intensity
including texture features.
Various SCOORD 2D outlines are illustrated in Figure 2. Note that for a POLYLINE, more than
two points are needed and it needs to be closed by using the same start and end points
30
.
Figure 2. Spatial Coordinates for Closed and Unclosed Geometric Objects
For example:
30
Earlier suggestions in [Clunie 2000] (predating [DICOM CP 233]) to use MULTIPOINT for this use case
are now inappropriate. Since CP 233, POLYLINE can and should be used.
SCOORD: (111041,DCM,"Outline") = POLYLINE
{17.8,27.1,13.8,34.0,15.2,24.3,17.8,27.1}
The POLYLINE definitions for SCOORD or SCOORD3D do not specify any winding order
31
. Nor
do they explicitly forbid crossing edges. It is hard to imagine a medical imaging use case for
which crossing edges are necessary, even though “holes” might be required (meaning that the
intersecting as opposed to actually crossing edges might be present, producing polygons that
are non-simple
32
; see section on Holes).
The POLYLINE encoding may be used regardless of whatever “tool” in the user interface may
have been used to create the line segments in the first place, or whatever visualization or
editing mechanism may be used subsequently. Specifically, whether the coordinates are
sequential points generated by a so-called “freehand” tool, as opposed to a tool defining
discrete points joined by line segments of non-trivial length, is not significant
33
.
Templates
There are two basic patterns commonly encountered when using closed objects and defined in
standard templates:
● A closed object illustrates a specific observation or a single measurement is made from
a closed object
● A closed object defines an ROI, from which one or more measurements may be made
In both cases, the basic mechanism for defining the object is exactly the same: the same forms
of SCOORD or SCOORD3D are used. The difference lies in the parent/child relationships of the
measurements or other observations, and the coordinates.
TID 1500 [TID1500] allows for both patterns, but encourages the definition of an ROI that is
defined graphically, and then contains one or more content items that convey the
measurements or the categorical observations. For example, using TID 1410 [TID1410], one
can encode
34
:
CONTAINER: (125007,DCM,"Measurement Group")
CONTAINS: SCOORD: (111030,DCM,"Image Region") = POLYLINE
{17.8,27.1,13.8,34.0,15.2,24.3,17.8,27.1}
31
The winding order is the relative order in which the vertex points of a polygon are listed, and may be
clockwise or counterclockwise when viewed from a particular direction [Norton 2014].
32
A simple polygon is a polygon that does not intersect itself and has no holes [WikiSimplePolygon].
33
The independence of the POLYLINE encoding from the authoring or editing tool begs the question of
how a recipient can select which tool to apply retrospectively if there is more than one choice. Since tools
are implementation-specific, and there is no widely accepted classification of such tools, the manner of
communicating the choice of tool is beyond the scope of the standard. Private codes or data elements
could be used to convey such information.
34
For clarity, this example omits the IMAGE child of SCOORD as well as some of the commonly used
TID 1410 content items such as (112039, DCM, "Tracking Identifier"), (121071, DCM, "Finding"),
(363698007, SCT, "Finding Site") and (370129005, SCT, "Measurement Method"). It also uses existing
categorical concepts and values used in the mammography CAD templates, such as (107644003, SCT,
"Shape").
CONTAINS: NUM: (42798000,SCT,"Area") = 40.1 (mm2,UCUM,"mm2")
CONTAINS: NUM: (81827009,SCT,"Diameter") = 6.3 (mm,UCUM,"mm")
CONTAINS: CODE: (107644003,SCT,"Shape") = (49608001,SCT,"Irregular")
In this example, the ROI is defined by spatial coordinates, and the numeric and categorical
assessments are enumerated without further coordinate references. I.e., the coordinates of the
ROI are considered sufficient. One implication of this, for example, is that the diameter of the
ROI is specified but the end points of whatever line segment was used to determine it are not
encoded.
Alternatively, using TID 1501 [TID 1501] within TID 1500 [TID 1500] and its invocation of TID
300 [TID 300], one could encode the same information without defining an ROI geometrically:
CONTAINER: (125007,DCM,"Measurement Group")
CONTAINS: NUM: (42798000,SCT,"Area") = 40.1 (mm2,UCUM,"mm2")
INFERRED FROM: SCOORD: = POLYLINE
{17.8,27.1,13.8,34.0,15.2,24.3,17.8,27.1}
CONTAINS: NUM: (81827009,SCT,"Diameter") = 6.3 (mm,UCUM,"mm")
INFERRED FROM: SCOORD: = POLYLINE {17.8,27.1,15.2,24.3}
CONTAINS: CODE: (107644003,SCT,"Shape") = (49608001,SCT,"Irregular")
In this example, the end points of the line segment for measuring the diameter are explicitly
specified, and no graphical information is provided for the categorical shape content item; all
being within the same Measurement Group, they can be assumed to apply to the same ROI
35
.
Separate Measurement Groups may be used:
CONTAINER: (125007,DCM,"Measurement Group")
CONTAINS: UIDREF: (112040,DCM,"Tracking Unique Identifier") = “1.2.3.8”
CONTAINS: NUM: (42798000,SCT,"Area") = 40.1 (mm2,UCUM,"mm2")
INFERRED FROM: SCOORD: = POLYLINE
{17.8,27.1,13.8,34.0,15.2,24.3,17.8,27.1}
35
You might well ask why the container concept name used is “Measurement Group” rather than “ROI” or
similar. Historically, templates that pre-date those used by TID 1500 [TID 1500] made extensive use of
containers generally referred to as “Measurement Groups”, using various concept names ranging from
the very general, such as (121070, DCM, "Findings"), to the very specific, such as (281231009, SCT,
"Blood Vessel of Head"), sometimes nested within other findings containers. However, other templates
required the grouping concept without a specific name, and (125007, DCM, “Measurement Group”) was
introduced in Supplement 26 OB-GYN Ultrasound Procedure Reports [Sup 26], for related measurements
and calculations that share a common context but without any other semantic implications. When the
planar and volumetric ROI templates were added by CP 1112 [DICOM CP 1112] and extended by CP
1386 [DICOM CP 1386], the grouping concept was reused, and inadvertently hardwired to the
Measurement Group concept code. In retrospect, it might have been better to use a concept like “ROI”,
since that was the explicit use case, or to allow it to be context specific, since the same templates can be
used for things that are arguably not ROIs per se, such as unclosed paths, and indeed categorical
observations without any measurements are permitted, but it is probably too late to relax the constraint
now.
CONTAINER: (125007,DCM,"Measurement Group")
CONTAINS: UIDREF: (112040,DCM,"Tracking Unique Identifier") = “1.2.3.8”
CONTAINS: NUM: (81827009,SCT,"Diameter") = 6.3 (mm,UCUM,"mm")
INFERRED FROM: SCOORD: = POLYLINE {17.8,27.1,15.2,24.3}
CONTAINER: (125007,DCM,"Measurement Group")
CONTAINS: UIDREF: (112040,DCM,"Tracking Unique Identifier") = “1.2.3.8”
CONTAINS: CODE: (107644003,SCT,"Shape") = (49608001,SCT,"Irregular")
In this example, the end points of the line segment for measuring the diameter are still explicitly
specified, and again no graphical information is provided for the categorical shape content item.
That they are of the same ROI is no longer implicit, but is conveyed by using the same Tracking
Unique Identifier.
The recommended approach for combining ROI definitions and measurements (such as area
and volume) with measurements that have additional geometric information (such as derived
linear dimensions) is to use the hybrid approach described in PS3.17 [PS3.17 RRR.5] and
discussed in the section on distance measurements from single line segments. Here is a similar
example, which defines an ROI geometrically, attributes to it measurements and observations
that are derived from the entire ROI, and separately describes those measurements associated
with the ROI but which have their own geometric information:
CONTAINER: (125007,DCM,"Measurement Group")
CONTAINS: UIDREF: (112040,DCM,"Tracking Unique Identifier") = “1.2.3.8”
CONTAINS: SCOORD: (111030,DCM,"Image Region") = POLYLINE
{17.8,27.1,13.8,34.0,15.2,24.3,17.8,27.1}
CONTAINS: NUM: (42798000,SCT,"Area") = 40.1 (mm2,UCUM,"mm2")
CONTAINS: CODE: (107644003,SCT,"Shape") = (49608001,SCT,"Irregular")
CONTAINER: (125007,DCM,"Measurement Group")
CONTAINS: UIDREF: (112040,DCM,"Tracking Unique Identifier") = “1.2.3.8”
CONTAINS: NUM: (81827009,SCT,"Diameter") = 6.3 (mm,UCUM,"mm")
INFERRED FROM: SCOORD: = POLYLINE {17.8,27.1,15.2,24.3}
Again note that the commonality of the Tracking Unique Identifier establishes that they are
measurements of the same ROI
36
.
The advantage of using the TID 1410 [TID1410] approach, to the extent possible, is particularly
apparent when very large numbers of observations are made on a single ROI, whether it be
through the automated generation of numerous quantitative texture features, or the selection of
a multitude of categorical descriptive elements.
Historically, the CAD and other application-specific templates in the standard have followed a
mixed approach, defining a region using coordinates and making multiple observations about it,
36
A subtlety in this example is that one Measurement Group actually defines the ROI geometrically,
whereas the other makes coordinate-based measurement upon it. This is not entirely obvious unless one
takes literally the “Image Region” concept name for the SCOORD that defines the ROI.
but then decorating size measurements with coordinates for them. In this example, using TID
4006 [TID 4006] invoking TID 4021 [TID 4021] and TID 1401 [TID 1401], one might encode:
CODE: (111059,DCM,"Single Image Finding") = (129769006,SCT,“Calcification
Cluster”)
HAS PROPERTIES: NUM: (111047,DCM,"Probability of cancer") = 90
(%,UCUM,"Percent")
HAS PROPERTIES: SCOORD: (111010,DCM,"Center") = POINT {15.6,28.4}
HAS PROPERTIES: SCOORD: (111041,DCM,"Outline") = POLYLINE
{17.8,27.1,13.8,34.0,15.2,24.3,17.8,27.1}
HAS PROPERTIES: CODE: (111009,DCM,"Calcification Type") =
(111344,DCM,“Fine pleomorphic calcification”)
HAS PROPERTIES: NUM: (42798000,SCT,"Area") = 40.1 (mm2,UCUM,"mm2")
INFERRED FROM: SCOORD: (121056,DCM,"Area Outline") = POLYLINE
{17.8,27.1,13.8,34.0,15.2,24.3,17.8,27.1}
Such an application-specific approach, which embodies an information model that is very
specific to the subject of interest, is only interoperable amongst dedicated applications. It may
be less able to be generalized, or to leverage generic implementations. A case in point is the
desire by contemporary developers to apply AI/ML techniques to mammography. They have a
requirement to encode and have displayed different quantitative measurements than anticipated
by the existing mammography CAD templates. The more general approach using TID 1500 is
inherently extensible; general rendering and extraction applications may be able to make use of
unexpected content encoded in expected places in an expected manner.
Bounding Boxes
Closed outlines and geometric objects do not always conform to the exact contours of a real
world object such as a tumor or other type of lesion. Two common use cases at opposite
extremes are to communicate the center or other approximate point indicator of the location of
an object, or to define a bounding box within which the object is entirely enclosed
37
. In neither
case are precise topographical or intensity based measurements likely to have much meaning.
The use of a Graphic Type of POINT to communicate an infinitely small such region has already
been discussed in the section on single points.
In practice, a geometric bounding object is typically encoded in exactly the same way as a
precise contour conforming to the edges of an object, except that geometric shapes rather than
contours are typically used. In such cases, there may be no explicit indication that an inexact
boundary is being encoded. Recently, a specific mechanism has been added [DICOM CP 1852]
to describe this use case. Here is an example of its use, without any accompanying
measurements:
CONTAINER: (125007,DCM,"Measurement Group")
37
This use case of a bounding region also supports the electronic equivalent of the traditional use of a
wax pencil to annotate locations on film or slide by hand drawing a rough circle around something of
interest. See also the discussion of arrow-like use cases.
CONTAINS: CODE: (130400,DCM,"Geometric purpose of region") =
(75958009,SCT,"Bounded by")
CONTAINS: SCOORD: (111030,DCM,"Image Region") = POLYLINE
{17.8,27.1,37.8,27.1,37.8,47.1,17.8,47.1,17.8,27.1}
It would probably be more useful if it were conveyed with some additional description, even if
there is no need for measurements, e.g., the type of finding:
CONTAINER: (125007,DCM,"Measurement Group")
CONTAINS: CODE: (121071,DCM,"Finding") = (86049000,SCT,"Neoplasm,
Primary")
CONTAINS: CODE: (130400,DCM,"Geometric purpose of region") =
(75958009,SCT,"Bounded by")
CONTAINS: SCOORD: (111030,DCM,"Image Region") = POLYLINE
{17.8,27.1,37.8,27.1,37.8,47.1,17.8,47.1,17.8,27.1}
Rectangles
A special case of a POLYLINE is the rectangle, as used in the bounding box example. A
rectangle obviously has certain constraints on the relationship between its vertices. Though
rectangles are very commonly used, there is no special graphic type for them, since defining
such would require definition of two vertices as well as an angle (since rectangles may be
rotated). It remains for the receiver of a POLYLINE object to recognize patterns of closed
polylines that are consistent with rectangles if for some reason it needs to know
38
.
Interpolation, Partial Pixels, Inclusion and Gaps
As discussed in the section on open multiple line segments, the standard does not define any
interpolation mechanism for objects defined by closed outlines, and in the absence of other
information, points should be assumed to be joined by straight line segments. This concern
does not apply to mathematically defined geometric objects.
The standard does not describe how to derive measurements from contours or geometric
objects, e.g., whether or not to account for partial pixel inclusion, or which pixels are included
depending on the method of interpolation, if any. However, since image-relative pixel locations
are specified with sub-pixel precision and 3D coordinates are relative to a frame of reference
and not a specific image, computations need to be performed without being quantized to any
arbitrary whole image pixel (or voxel) boundaries
39
.
The image-relative sub-pixel coordinates are defined such that zero represents one edge of a
pixel (or voxel), 0.5 the middle of it and 1.0 the other edge (being the same as the edge of the
38
For instance, in order to render a rectangle on a display using optimized graphics primitives that draw
rectangles in a more specific manner than arbitrary polygons.
39
Even though the source images or volumes themselves are by definition so quantized, or may have
been resampled into some other quantization representation.
adjacent pixel). Hence an unrotated rectangle with coordinates of (0.0,0.0, 1.0,0.0, 1.0, 1.0,
0.0,1.0, 0.0,0.0) is expected to have an area of exactly 1 pixel2, i.e., has the area fully contained
within the closed object with an infinitely thin edge, irrespective of the pixel boundary alignment.
Translating this rectangle by half a pixel in any direction will result in exactly the same derived
area
40
. This aligns with the interpretation of the mathematically defined geometric objects
(circles and ellipses and ellipsoids, which have algebraically-defined derived measurements), as
well as the use of 3D objects with non-image-relative coordinates. To be consistent with that
interpretation, when lines traverse a pixel, rather than lie exactly on its boundary
41
, then partial
occupancy is inherently accounted for. Or to put it another way, the alignment with pixel
boundaries is completely irrelevant to the calculation
42
.
The standard does not require or suggest any particular sampling rate. For example, there is no
requirement that the coordinates be restricted to exact pixel or voxel locations, or the centers
thereof in sub-pixel resolution. For image-relative coordinates (SCOORD) the in-plane sampling
rate of the coordinates may be different from the pixel spacing of the image. The cross-plane
sampling rate may be more coarse than for the image slices since a slice reference is required.
There is no requirement that the sampling be regular, in-plane or between planes. However, if
an ROI is not encoded on every slice, the question arises as to whether those slices are
expected to be omitted from the ROI or included. This is not defined by the inherent coordinate
mechanism, but may be explicitly specified in the appropriate template used. TID 1411 specifies
that the entire spatial extent of the ROI is included, regardless of whether a contour on every
intermediate slice is present, and that fragmented objects need to be described as separate
ROIs, with separate identifiers but other common metadata such as the Finding code
43
. There
are essentially no restrictions on the SCOORD3D sampling rate, nor for that matter, are planar
3D coordinate based structures required to be in the same plane as any particular (set of) image
slices.
Computed values could be different if the path were to be interpolated rather than using straight
line segments. Indeed if the shape represented by the path is sparsely sampled, which image
40
But it will not produce the same value for intensity rather than size related measurements, if the
covered adjacent pixels have different intensities. Accounting for this correctly is an important factor for
achieving repeatable measurements using different implementations.
41
The question of edge points is specifically mentioned in DICOM RTSS, where it is stated that “points in
space lying along the path defined by the contour are considered to be part of the ROI” [PS3.3 C.8.8.6.3].
However, since RTSS coordinates are 3D and not image-relative, it is not clear what this is intended to
mean, since a point in 3D space is infinitely small and a line defined in 3D space is infinitely thin.
42
Alignment to pixel boundaries may not have been irrelevant to the tool that was used to generate the
contour in the first place. For example, such a tool that does not support sub-pixel precision, and which
aligns hand drawn contours with entire pixels, and which does not account for partial area/volume effects
when computing derived measurements, will record a numerically different result compared to that
derived by ignoring the pixel quantization and using the serialized coordinates, which should have been
centered on the pixels if correctly encoded (with “0.5” values).
Further, other conventions may have been defined for the source of contours that are being converted.
For example, unusually, the LIDC XML format for describing lung nodules specifies voxel-aligned
coordinates that are outside the ROI (“first voxel just outside the nodule”) and are not themselves to be
included [LIDC XML 2010] [Fedorov 2020].
43
This clarification was added in CP 2044, prior to which it was undefined.
pixels are included at all may be different, but as stated above, which pixels are included is not
a factor. Given that interpolation is not specified by the standard, repeatable computations can
only be obtained by using straight line segments.
Holes
DICOM SR has no infrastructural elements to explicitly convey the concept of a “hole” in a
closed contour or geometric object, i.e., regions to exclude from the enclosing region
44
.
Yet in real life this may be a useful scenario for creating a detailed description. One may want to
describe characteristics or measurements of the enhancing rim of a tumor, for example, or
exclude central necrotic or cystic regions. At a microscopic level one might want to describe the
cytoplasm, excluding the nucleus or other organelles
45
.
At present, therefore, the semantics of the concepts encoded in the content items must suffice
to describe regions that are enclosed within other regions. I.e., the recipient must understand
(have a sufficient ontology to recognized that), for example, a nucleus is
Historically, in RTSS, holes referred to as “excluded inner volumes” have been theoretically
possible by using a “keyhole technique”, encoding an arbitrarily narrow channel connecting the
outer contour to the inner contour, so that it is drawn as a single contour within one set of
coordinates [PS3.3 C.8.8.6.3]
4647
. Conceivably, this mechanism could be repeated to allow for
multiple nested included and excluded regions
48
. It is not known how widely implemented this
mechanism is for RTSS, or what problems it presents for analysis, measurement computation
and rendering. The DICOM SR SCOORD POLYLINE and SCOORD3D POLYLINE and
POLYGON descriptions do not explicitly forbid this representation, but it would likely be
regarded as a non-interoperable abuse if it were to be used without calling attention to its exact
interpretation from within the standard.
44
Standards from the Geographic Information Systems (GIS) world do consider holes explicitly. See, for
example, the “interior ring” concept from GeoJSON [GeoJSON RFC], which allows one level of multiple
holes. GeoJSON polygons are encoded as an array of “linear rings” each specified as a nested
coordinate array, the first (or only) of which is the “exterior ring” and the rest are “interior rings” to be
excluded. More recent versions of GeoJSON also specify a right hand winding rule for polygons, a
restriction that DICOM does not require.
45
Such histopathology use cases are currently the subject of exploratory work in the DICOM WG 26
Pathology ad hoc subgroup on Annotations.
46
This capability was added in CP 439 [DICOM CP 439], and modified and extended in CP 2037 [DICOM
CP 2037], which also allowed for components within an inner void.
47
Such intersecting contours make the resulting closed polygon non-simple, even though the contours
are not actually crossing.
48
There is an issue with more than one level of nesting. The DICOM RTSS definition states that “points in
space lying along the path defined by the contour are considered to be part of the ROI”. This is OK for the
arbitrarily thin connection between the connection of the first outer contour to inner contour, since it
traverses the region between them, which is within the included region. However, the next connection, to
the outer contour of an enclosed region (within the first excluded region) traverses an excluded region but
the points it passes through would be included.
Indication of Location, Direction and Magnitude – Arrow-like Things
DICOM SR does not have any graphic primitives representing “arrows”, such as a human might
draw on an image to “point to something”, since DICOM SR annotations are intended to be
semantically meaningful in a precise and specific way. They are also intended to be
interpretable (especially by a machine) without requiring out of band information, or the sort of
inference that currently requires a human.
Many use cases for an arrow where the subject of interest is at or near
49
the tip of the arrow can
be addressed in the more semantically meaningful manner described for a single point, e.g.:
SCOORD: (111010,DCM,"Center") = POINT {128.5,128.5}
Whether a receiving application chooses to render this as a point, cross-hair, tip of an arrow or
some other geometric construct is beyond the scope of DICOM SR to define
50
.
For most use cases, the direction and length of the arrow are irrelevant, since the subject matter
is at its tip. In some cases, the arrow tip may be placed only near the object of interest rather
than directly on it. One might explicitly address the imprecision of the location as follows:
SCOORD: (121226,DCM,"Approximate spatial location") = POINT {128.5,128.5}
which could always be used by an application that was never sure how close the arrow tip was
likely to be.
When either direction and/or magnitude are actually important, then a single two-point
POLYLINE can be used. For example:
SCOORD: (121225,DCM,"Vector") = POLYLINE {17.8,27.1,13.8,34.0}
More information about the meaning of the location, direction
51
or magnitude may be apparent
either from the context of use of the SCOORD or its concept name (which may be absent). For
example, one can use to [TID 1410] to encode the location of a primary tumor located at the tip
of what the user might have indicated with an arrow:
CONTAINER: (125007,DCM,"Measurement Group")
49
A user may have drawn an arrow that is only near an indicated structure so that its rendering does not
obscure that structure; this is semantically undesirable but common practice.
50
Just as what color, line thickness and style, font, weight or any other purely rendering related issue is
deemed to be beyond the scope of SR.
51
Strictly speaking, since DICOM does not currently specify that the order that the points in a POLYLINE
are encoded is the order in which they are to be interpreted. I.e., the assumptions that the direction is
from the first towards the second point, or that the second point is the tip of an arrow rather than the first
point, may be unsound. A new CP is proposed to clarify that when direction is semantically important, it is
from earlier to later points.
CONTAINS: CODE: (121071,DCM,"Finding") = (86049000,SCT,"Neoplasm,
Primary")
CONTAINS: SCOORD: (111030,DCM,"Image Region") = POINT {17.8,27.1}
SELECTED FROM: IMAGE: = (CT Image,1.2.3.4)
Or one can use [TID 1501] to describe a location and a feature without establishing an ROI or
making a measurement
52
, as follows:
CONTAINER: (125007,DCM,"Measurement Group")
CONTAINS: SCOORD: (86049000,SCT,"Neoplasm, Primary") = POINT {17.8,27.1}
SELECTED FROM: IMAGE: = (CT Image,1.2.3.4)
When faced with these use cases, implementers often assert that their users demand
reproducibility of appearance, as well as editability of that appearance after initial storage (i.e.,
for serialization and deserialization of the same tool or widget). One standard solution to this
problem of mixing semantics and appearance
53
is the use of a Presentation State (PR) object to
accompany the SR
54
. Each IMAGE content item may contain a reference (in a nested
Referenced SOP Sequence (0008,1199)) to such a PR instance, whose Graphic Annotation
Sequence (0070,0001) can describe the visual appearance
55
.
Angles – With or Without a Shared Vertex
DICOM SR does not have any graphic primitives intended specifically for measuring angles.
Accordingly, it is necessary to use one or more of the existing primitives, such as POLYLINE, in
conjunction with coded concepts signaling the purpose of the graphic entity.
Since line segments are involved, the POLYLINE graphic type may be used to define those line
segments
56
the locations of their end points. Both 2D and 3D points are supported.
52
CP 1999 [DICOM CP 1999] extended TID 1501 to allow image, waveform and coordinate references
despite the absence of a measurement.
53
Rather than inappropriately trying to use SR POLYLINE SCOORD content items merely as rendering
tools, e.g., to draw an arrowhead.
54
Non-standard solutions typically involve the addition of private data elements to preserve editable and
visual features not encoded in standard SR constructs. The weaknesses of this approach should be
obvious.
55
This is a relatively heavy weight solution, but is possible. One down side, theoretically, is that the entire
grayscale and graphic rendering pipeline (including windowing and zoom/pan) are described by the PR
instance, but arguably they could be ignored by the recipient for this use case of rendering the SR, rather
than the PR per se.
56
Earlier suggestions for describing angles in DICOM SR using MULTIPOINT (e.g., in [Clunie 2000])
were motivated by the original restriction of POLYLINE to being only closed polygons. This restriction has
subsequently been removed [DICOM CP 233], so the POLYLINE now appears to be the most appropriate
choice. It has the significant benefit that when rendered literally by a naive application that is unaware of
the angle-related semantics, it will appear as two line segments, which is visually more meaningful than a
set of unrelated dots.
Angles defined by the intersection of two line segments may, within an image (or scope of a 3D
volume):
● share a common vertex (Figure 3), or
● not share a common vertex but define an angle when extended (Figure 4)
57
Figure 3. Angle with Shared Vertex
Figure 4. Angle without Shared Vertex
A shared vertex angle can be represented by a single POLYLINE SCOORD content item, in
which the shared vertex is the second of two endpoints. For example:
57
The Cobb angle [Weerakkody], used for measuring the curvature of the spine by the angle between
line segments drawn along the endplates of the vertebra is a classic example of this use case and is
implemented in many radiology viewers [ClearCanvas ShowAngles] [Cornerstone.js CobbAngle].
NUM: (121224,DCM,"Acetabular angle") = 28 (deg,UCUM,"degrees")
SCOORD: (110859,DCM,"Angle") = POLYLINE {9.5,9.5,0.5,9.5,8.4,5.3}
In this example, the concept (110859, DCM, "Angle")
58
is encoded as the concept name of the
SCOORD content item, even though it is not required, in order to signal that it is an angle, since
the parent concept name of the NUM measurement may be highly application-specific
59
.
An angle without a shared vertex can be represented by a pair of POLYLINE SCOORD content
items with the concept of each SCOORD content item explicitly identified as representing the
arm of an angle
60
. For example:
NUM: (285285000,SCT,"Cobb angle") = 28 (deg,UCUM,"degrees")
SCOORD: (121223,DCM,"Arm of angle") = POLYLINE {9.5,9.5,0.5,9.5}
SCOORD: (121223,DCM,"Arm of angle") = POLYLINE {5.5,6.5,13.4,2.3}
Obviously, the shared vertex angle can be handled with the same two content item approach,
and the receiving application can be left to detect that the vertex is shared, if it matters.
Rewriting the acetabular angle example (with the direction of one of the line segments reversed
so that the common vertex is the first point of both line segments)
61
, we have:
NUM: (121224,DCM,"Acetabular angle") = 28 (deg,UCUM,"degrees")
SCOORD: (121223,DCM,"Arm of angle") = POLYLINE {0.5,9.5,9.5,9.5}
SCOORD: (121223,DCM,"Arm of angle") = POLYLINE {0.5,9.5,8.4,5.3}
Anonymous Image and Coordinate Content Items
As discussed in the description of the IMAGE, SCOORD and SCOORD3D content items, the
concept name (purpose of reference) is optional for spatial coordinates and images
62
. The intent
is that when an image reference occurs without any coordinates, then the purpose of reference
may be sent as the concept name of the image content item; indeed it probably should. When
coordinates are selected from an image, however, it makes more sense to denote the purpose
of reference with the coordinate node. This is because the image reference itself has no
“purpose” per se, other than to act as the target of the coordinates. The “concept” being
identified is defined by the coordinates applied to the image, not the image itself. So images that
are the targets of SELECTED FROM relationships with coordinates usually do not have a
concept name.
58
The DCM code 110589 is used here rather than the SNOMED CT code 1483009, which actually
means “angular (qualifier value”), even though it has a synonym of “angle”.
59
Acetabular angle is a well-defined orthopedic measurement for assessing developmental dysplasia of
the hip (DDH). See [Gaillard].
60
The DCM code 121223 was added by CP 2000 for this purpose, and its use to define angles without
shared vertices explicitly described in the corresponding TID 1500 invoked sub-templates.
61
There is no standard requirement or particular reason to do so, however.
62
It is also optional for COMPOSITE and WAVEFORM content items for the same reason.
Similarly, when an image reference or a set of coordinates (whether selected from an image or
not) is used as the basis of the definition or source of a higher level entity, such as a region of
interest or a numeric measurement or a categorical statement, again it may be appropriate to
omit the concept names of the image or coordinate content items, since the purpose of
reference is described with the higher level entity.
In this example using the TID 1500 template, neither the image nor the coordinate references
have explicit purposes of references (i.e., the content items have no concept names).
CONTAINER: (,,"Measurement Group")
HAS CONCEPT MOD: CODE: (,,"Finding Site") = (,,"Kidney")
CONTAINS: NUM: (,,"Length") = 66.43856134 (,,"mm")
INFERRED FROM: SCOORD: = POLYLINE {17.8,27.1,13.8,34.0}
SELECTED FROM: IMAGE: = (CT Image,1.2.3.4)
The purpose of the SCOORD (to define the end points of a length measurement) is implicit in
being the child of the parent NUM content item, which has a concept name of (410668003,
SCT, "Length"). Further, that its purpose is to measure the length of the kidney, is implicit in the
NUM content item having a sibling defining the (363698007,SCT,"Finding Site") as being
(64033007,SCT,"Kidney").
In general, most existing SR templates follow the pattern of using anonymous content items for
image and coordinate references, except when it is necessary to distinguish the reference from
other siblings, such as:
● to define the region for a finding,
● to illustrate rather than define a finding,
● to distinguish an image that was used for segmentation rather than the segmentation
itself, or
● to visually explain an observation.
For example, using TID 1410 [TID 1410] from TID 1500 [TID 1500], one might encode:
SCOORD: (,,"Image Region") = POLYLINE {17.8,27.1,13.8,34.0}
SELECTED FROM: IMAGE: = (CT Image,1.2.3.4)
IMAGE: (,,"Illustration of ROI") = (SC Image,1.2.3.5)
or
IMAGE: (,,"Referenced Segmentation Frame") = (SEG Image,1.2.3.6)
IMAGE: (,,"Source image for segmentation") = (CT Image,1.2.3.4)
IMAGE: (,,"Visual explanation") = (Parametric Map Image,1.2.3.7)
This is not to say that an implementer cannot send a concept name for an image or coordinate
content item, even if the template does not require it, only that the receiver may not depend on
its presence. E.g., one could encode the earlier example as:
CONTAINER: (,,"Measurement Group")
HAS CONCEPT MOD: CODE: (,,"Finding Site") = (,,"Kidney")
CONTAINS: NUM: (,,"Length") = 66.43856134 (,,"mm")
INFERRED FROM: SCOORD: (,,"Path") = POLYLINE {...}
SELECTED FROM: IMAGE: (,,"Source") = (...)
In this case, (121055, DCM, “Path”) is the closest thing to a generic line segment descriptor that
is currently present in the standard
63
. The concept (260753009, SCT, “Source”) is a generic
attribute in SNOMED CT
64
. Another option might be to use (121112, DCM, “Source of
Measurement”) for either the SCOORD content item
65
or the IMAGE content item.
Tracking of Planar Annotations
Tracking Unique Identifier and Tracking Identifier
The use of (112039, DCM, "Tracking Identifier") and (112040, DCM, "Tracking Unique
Identifier") has already been illustrated for several examples of specific types of annotations, but
general features will be summarized here. These comments are not restricted to planar
annotations, since volumetric and other types of annotations are handled similarly.
In short, ROIs defined in one Measurement Group, whether it be within a single SR Instance or
different SR Instances, may be identified as being of the same entity by commonality of the
Tracking Unique Identifier, which is globally unique.
For example, a single Target lesion detected at a baseline CT study, may be correlated with the
appearance of the same lesion on a later study (later “time point”) by using the same value for
Tracking Unique Identifier. This is despite the fact that the images being referenced and indeed
the physical characteristics of the lesion will have changed.
The (112039, DCM, "Tracking Identifier") is intended to be a human-readable correlate of the
Tracking Unique Identifier, and need only be unique within an appropriate context. E.g., it might
be “Target Lesion 001” or similar, unique only within the scope of a single set of reports by a
single reader about a single patient.
The Tracking Unique Identifier is intended to reference the same physical entity, regardless of
how it is imaged or measured. So, for example, if the same lesion were examined with both CT,
PET and MR, ideally the same Tracking Unique Identifier would be used for all ROIs regardless
of modality.
63
Indeed, (121055, DCM, “Path”) is required as the concept name for the SCOORD used in TID 1400
Linear Measurement [TID 1400], a template that is not invoked from TID 1500 but is used in
Mammography CAD.
64
The suggestion to use (260753009, SCT, “Source”) was added by CP 2036.
65
As illustrated in [Clunie 2011], and used extensively in the standard for waveform measurements.
Quite complex patterns of lesion tracking are encountered in practice, and include such
scenarios as splitting and merging of lesions [Clunie 2007]. For such cases, different Tracking
Unique Identifiers would generally be assigned, and there are no standard templates for
defining their relationships at this time, though some preliminary work on encoding such
relationships in DICOM SR has been documented [Clunie 2011].
The Tracking Unique Identifier and Tracking Identifier may also be used in other DICOM
Storage SOP Class Instances, such as Segmentations. So, for example, the relationship
between an entity described by an ROI in an SR may be associated with the same entity in a
Segmentation, whether or not there is an explicit reference from the SR to the Segmentation.
There is no requirement to include the identifiers, however, and an explicit reference may be
sufficient.
It is understood that correlation of lesions across time, modalities, studies or even within the
same study requires software dedicated to the purpose or facilities that allow the users to
establish the commonality. Accordingly, there may well be occasions on which the information is
not available and either the Tracking Unique Identifier is absent or it has different values in
different Measurement Groups or SR Instances, indefinitely or until commonality is established
and a new SR can be issued.
Observation UID
All DICOM SR Content Items may have a value for the Observation UID (0040,A171) Attribute.
This is defined to be a unique identifier of “the semantic content … of the observation … and its
subsidiary Content Items, if any”. Furthermore, the same UID may be used in different
representations (e.g., an HL7 CDA equivalent of a DICOM SR).
Though rarely used, this is more generic than the Tracking Unique Identifier, in that it can apply
to any entity. Conversely, it is more specific, in that it refers to one observation of an entity, not
the same entity on different occasions.
One potential use for this is to permit external references to an observation about an entity
defined within an SR. For example, if the same planar annotation was used in two different SR
Instances, and the content of that annotation was identical (same metadata and coordinates
and image references), then in theory, the same Observation UID could be used in both SR
Instances. There is no documented experience using this approach, but it is theoretically
possible.
Another potential use is to associate Presentation States with partial SR content. Specific
Attributes that reference Observation UID were added in the Volumetric Graphic Annotation
Module of the Volumetric Presentation States for this purpose, but there is again no experience
with the actual utility of this.
Referenced Content Items
The foregoing use cases are to be distinguished from those in which there is a reference from
one part of an SR instance to another part of an instance.
There is a specific mechanism to support this, the so-called “by-reference” relationship and the
Referenced Content Item Identifier (0040,DB73). The successive numeric values of this
Attribute denote a path descending from the root of the tree to the referenced node. This
establishes a mechanism to represent a directed acyclic graph (DAG) rather than only a tree.
The intention of this mechanism is to avoid the need to replicate content when it needs to be
used more than once, as well as to avoid having to establish commonality of identity of
replicated content.
A classic example is when a measurement is derived from, or inferred from, one or more other
measurements in the same instance.
Though it appears attractive, this approach creates problems for recipients trying to understand,
represent, tabulate or otherwise visualize the content. Within a sophisticated rendering, the
references might be managed as hyperlinks, but the general case is challenging. Successful
implementations may need to be use case specific.
The Mammo CAD DICOM SR templates, for example, de-referenced all of their image
references to the more detailed description of each image in the Image Library. In retrospect, it
might have been considerably simpler for recipients if each IMAGE content item had been
duplicated in place, and on the rare occasions when the Image Library entry was needed, it
could have been selected by the commonality of the referenced SOP Instance UID of the image
in question.
The general purpose template TID 1500 allows for derived measurements to reference their
source measurements using this mechanism, but otherwise makes very little use of by-
reference relationships, and authors are encouraged to avoid them, if possible.
Conclusions
Though the DICOM SR framework is very generic, the use of specific patterns and
templates can constrain creativity when implementing common use cases to achieve a high
level of application interoperability. This applies not only to creating, extracting, tabulating
and visualizing content, but also to editing it, and potentially using DICOM SR as the single
format for serializing and de-serializing planar annotations.
Interoperability also requires the use of standard codes, to the extent that they are defined,
since the choice of code may have low level semantic as well as structural significance,
beyond merely defining the quantity being measured. In some cases private codes will be
needed for exotic or novel quantities and metrics, but in all cases standard codes and the
templates and value sets (context groups) that are available should be used.
Most if not all planar annotations commonly generated by workstations can be encoded
using SR as described. In some cases, common practice may not be the best practice (e.g.,
the use of arrows), yet it is necessary to take advantage of the installed base of systems
and archived objects, even if they have limited utility. That said, going forward with new
systems, improvements to existing systems and for new objects and new protocols defining
what new objects are to be captured, semantically rigorous best practices for SR encoding
of planar annotations should be followed.
References
[Wiki ROI] http://en.wikipedia.org/wiki/Region_of_interest
[DICOM SEG] National Electrical Manufacturers Association (NEMA). Digital Imaging and
Communications in Medicine (DICOM) Standard PS3.3 - Information Object Definitions - A.51
Segmentation IOD. Rosslyn, VA: National Electrical Manufacturers Association (NEMA).
Available from: http://dicom.nema.org/medical/dicom/current/output/chtml/part03/sect_A.51.html
[DICOM SEG ref from TID 1410] National Electrical Manufacturers Association (NEMA). Digital
Imaging and Communications in Medicine (DICOM) Standard PS3.16 - Content Mapping
Resource - TID 1410 Planar ROI Measurements and Qualitative Evaluations - Row 7 -
Referenced Segmentation Frame.
http://dicom.nema.org/medical/dicom/current/output/chtml/part16/chapter_A.html#para_1d5a0bc
4-a020-44ab-a211-f17a9fa83246
[SCOORD] National Electrical Manufacturers Association (NEMA). Digital Imaging and
Communications in Medicine (DICOM) Standard PS3.16 - Content Mapping Resource - C.18.6
Spatial Coordinates Macro. Rosslyn, VA: National Electrical Manufacturers Association (NEMA).
Available from:
http://dicom.nema.org/medical/dicom/current/output/chtml/part03/sect_C.18.6.html
[SCOORD3D] National Electrical Manufacturers Association (NEMA). Digital Imaging and
Communications in Medicine (DICOM) Standard PS3.16 - Content Mapping Resource - C.18.9
3D Spatial Coordinates Macro. Rosslyn, VA: National Electrical Manufacturers Association
(NEMA). Available from:
http://dicom.nema.org/medical/dicom/current/output/chtml/part03/sect_C.18.9.html
[DICOM PR] National Electrical Manufacturers Association (NEMA). Digital Imaging and
Communications in Medicine (DICOM) Standard PS3.3 - Information Object Definitions - A.33
Softcopy Presentation State Information Object Definitions. Rosslyn, VA: National Electrical
Manufacturers Association (NEMA). Available from:
http://dicom.nema.org/medical/dicom/current/output/chtml/part03/sect_A.33.html
[DICOM Conceptual Volumes] National Electrical Manufacturers Association (NEMA). Digital
Imaging and Communications in Medicine (DICOM) Standard PS3.3 - Information Object
Definitions - 10.34 Conceptual Volume Segmentation Reference and Combination Macro.
Rosslyn, VA: National Electrical Manufacturers Association (NEMA). Available from:
http://dicom.nema.org/medical/dicom/current/output/chtml/part03/sect_10.34.html
[DICOM RTSS] National Electrical Manufacturers Association (NEMA). Digital Imaging and
Communications in Medicine (DICOM) Standard PS3.3 - Information Object Definitions - A.19
RT Structure Set IOD. Rosslyn, VA: National Electrical Manufacturers Association (NEMA).
Available from: http://dicom.nema.org/medical/dicom/current/output/chtml/part03/sect_A.19.html
[Clunie 2000] Clunie DA. DICOM Structured Reporting. PixelMed Publishing; 2000. Available
from: http://pixelmed.com/srbook.html
[Clunie 2007] Clunie DA. DICOM Structured Reporting and Cancer Clinical Trials Results.
Cancer Informatics. 2007 May 12;4(CIN-ImSI-Clunie-et-al). Available from:
http://journals.sagepub.com/doi/10.4137/CIN.S37032
[Clunie 2011] Clunie DA. Encoding of Oncology Clinical Trial Read Results in DICOM SR -
Abbreviated QIBA 1C Content. 2011. Available from:
http://qibawiki.rsna.org/images/f/fc/ClinicalTrialResults_20111123_QIBA.doc
[Fischer 2018] Fischer J, Fedorov A, Pieper S, Clunie D, Kast J. Mint Lesion planar annotations
and measurements. 2018. Available from: https://docs.google.com/document/d/1TA-
iGEWl42hAtKtBMvfgsHdud3OMlrl-hkWN3uG1ZRM
[TID 1500] National Electrical Manufacturers Association (NEMA). Digital Imaging and
Communications in Medicine (DICOM) Standard PS3.16 - Content Mapping Resource - TID
1500 Measurement Report. Available from:
http://dicom.nema.org/medical/dicom/current/output/chtml/part16/chapter_A.html#sect_TID_150
0
[TID 1410] National Electrical Manufacturers Association (NEMA). Digital Imaging and
Communications in Medicine (DICOM) Standard PS3.16 - Content Mapping Resource - TID
1410 Planar ROI Measurements and Qualitative Evaluations. Available from:
http://dicom.nema.org/medical/dicom/current/output/chtml/part16/chapter_A.html#sect_TID_141
0
[TID 300] National Electrical Manufacturers Association (NEMA). Digital Imaging and
Communications in Medicine (DICOM) Standard PS3.16 - Content Mapping Resource - TID 300
Measurement. Available from:
http://dicom.nema.org/medical/dicom/current/output/chtml/part16/chapter_A.html#sect_TID_300
[TID 320] National Electrical Manufacturers Association (NEMA). Digital Imaging and
Communications in Medicine (DICOM) Standard PS3.16 - Content Mapping Resource - TID 320
Image or Spatial Coordinates. Available from:
http://dicom.nema.org/medical/dicom/current/output/chtml/part16/chapter_A.html#sect_TID_320
[dicom3tools] Clunie D. Dicom3tools Software. Available from:
http://www.dclunie.com/dicom3tools.html
[Cross-Sectional Image Library] National Electrical Manufacturers Association (NEMA). Digital
Imaging and Communications in Medicine (DICOM) Standard PS3.16 - Content Mapping
Resource - TID 1604 Image Library Entry Descriptors for Cross-Sectional Modalities. Available
from:
http://dicom.nema.org/medical/dicom/current/output/chtml/part16/chapter_A.html#sect_TID_160
4
[TID 4021] National Electrical Manufacturers Association (NEMA). Digital Imaging and
Communications in Medicine (DICOM) Standard PS3.16 - Content Mapping Resource - TID
4021 Mammography CAD Geometry. Available from:
http://dicom.nema.org/medical/dicom/current/output/chtml/part16/sect_TID_4021.html
[TID 4107] National Electrical Manufacturers Association (NEMA). Digital Imaging and
Communications in Medicine (DICOM) Standard PS3.16 - Content Mapping Resource - TID
4107 Chest CAD Geometry. Available from:
http://dicom.nema.org/medical/dicom/current/output/chtml/part16/sect_TID_4107.html
[TID 4129] National Electrical Manufacturers Association (NEMA). Digital Imaging and
Communications in Medicine (DICOM) Standard PS3.16 - Content Mapping Resource - TID
4129 Colon CAD Geometry. Available from:
http://dicom.nema.org/medical/dicom/current/output/chtml/part16/sect_TID_4129.html
[DICOM CP 1999] DICOM Standards Committee. DICOM CP 1999 - Non-measurement 2D and
3D coordinate and image references. National Electrical Manufacturers Association (NEMA);
2019. Available from:
ftp://medical.nema.org/medical/dicom/final/cp1999_ft_TID1501NonMeasurementCoordImageRe
ferences.pdf
[DICOM CP 1112] DICOM Standards Committee. DICOM CP 1112 - Add SR Structure for ROI
Measurements. National Electrical Manufacturers Association (NEMA); 2012. Available from:
ftp://medical.nema.org/medical/dicom/final/cp1112_ft.pdf
[DICOM CP 1386] DICOM Standards Committee. DICOM CP 1386 - Addition of Measurement
Report Root Template for Planar and Volumetric ROIs. National Electrical Manufacturers
Association (NEMA); 2014. Available from:
ftp://medical.nema.org/medical/dicom/final/cp1386_ft.pdf
[DICOM CP 1789] DICOM Standards Committee. DICOM CP 1789 - Allow categorical
observations without measurements for ROIs in TID 1500. National Electrical Manufacturers
Association (NEMA); 2018. Available from:
ftp://medical.nema.org/medical/dicom/final/cp1789_ft_TID1500allowcategoricalalone.pdf
[TID 4006] National Electrical Manufacturers Association (NEMA). Digital Imaging and
Communications in Medicine (DICOM) Standard PS3.16 - Content Mapping Resource - TID
4006 Mammography CAD Single Image Finding. Available from:
http://dicom.nema.org/medical/dicom/current/output/chtml/part16/sect_TID_4006.html
[TID 1501] National Electrical Manufacturers Association (NEMA). Digital Imaging and
Communications in Medicine (DICOM) Standard PS3.16 - Content Mapping Resource - TID
1501 Measurement and Qualitative Evaluation Group. Available from:
http://dicom.nema.org/medical/dicom/current/output/chtml/part16/chapter_A.html#sect_TID_150
1
[CID 7470] National Electrical Manufacturers Association (NEMA). Digital Imaging and
Communications in Medicine (DICOM) Standard PS3.16 - Content Mapping Resource - CID
7470 Linear Measurements. Available from:
http://dicom.nema.org/medical/dicom/current/output/chtml/part16/sect_CID_7470.html
[CID 218] National Electrical Manufacturers Association (NEMA). Digital Imaging and
Communications in Medicine (DICOM) Standard PS3.16 - Content Mapping Resource - CID
218 Quantitative Image Features. Available from:
http://dicom.nema.org/medical/dicom/current/output/chtml/part16/sect_CID_218.html
[Sup 26] DICOM Standards Committee. DICOM Supplement 26 - OB-GYN Ultrasound
Procedure Reports. National Electrical Manufacturers Association (NEMA); 2003. Available
from: ftp://medical.nema.org/medical/dicom/final/sup26_ft.pdf
[PS3.17 RRR.5] National Electrical Manufacturers Association (NEMA). Digital Imaging and
Communications in Medicine (DICOM) Standard PS3.17 - Explanatory Information - on CT
Example. Available from:
http://dicom.nema.org/medical/dicom/current/output/chtml/part17/sect_RRR.5.html
[TID 5016] National Electrical Manufacturers Association (NEMA). Digital Imaging and
Communications in Medicine (DICOM) Standard PS3.16 - Content Mapping Resource - TID
5016 LWH Volume Group. Available from:
http://dicom.nema.org/medical/dicom/current/output/chtml/part16/sect_TID_5016.html
[TID 1420] National Electrical Manufacturers Association (NEMA). Digital Imaging and
Communications in Medicine (DICOM) Standard PS3.16 - Content Mapping Resource - TID
1420 Measurements Derived From Multiple ROI Measurements. Available from:
http://dicom.nema.org/medical/dicom/current/output/chtml/part16/chapter_A.html#sect_TID_142
0
[TID 3214] National Electrical Manufacturers Association (NEMA). Digital Imaging and
Communications in Medicine (DICOM) Standard PS3.16 - Content Mapping Resource -
Analyzed Segment Analyzed Segment. Available from:
http://dicom.nema.org/medical/dicom/current/output/chtml/part16/sect_TID_3214.html
[TID 4008] National Electrical Manufacturers Association (NEMA). Digital Imaging and
Communications in Medicine (DICOM) Standard PS3.16 - Content Mapping Resource - TID
4008 Mammography CAD Breast Geometry. Available from:
http://dicom.nema.org/medical/dicom/current/output/chtml/part16/sect_TID_4008.html
[Norton 2014] Norton J. Determining the Winding of a Polygon Given as a Set of Ordered
Points. 2014. Available from: http://www.element84.com/blog/determining-the-winding-of-a-
polygon-given-as-a-set-of-ordered-points
[WikiSimplePolygon] Simple polygon. In: Wikipedia. 2019. Available from:
http://en.wikipedia.org/wiki/Simple_polygon
[DICOM PS3.3 C.10.5.1.2] National Electrical Manufacturers Association (NEMA). Digital
Imaging and Communications in Medicine (DICOM) Standard PS3.3 - Information Object
Definitions - C.10.5.1.2 Graphic Data and Graphic Type. Rosslyn, VA: National Electrical
Manufacturers Association (NEMA). Available from:
http://dicom.nema.org/medical/dicom/current/output/chtml/part03/sect_C.10.5.html#sect_C.10.5.
1.2
[DICOM CP 233] DICOM Standards Committee. DICOM CP 233 - Clarify SR closed polyline
start and end points. National Electrical Manufacturers Association (NEMA); 2001. Available
from: ftp://medical.nema.org/medical/dicom/final/cp233_ft.pdf
[DICOM CP 216] DICOM Standards Committee. DICOM CP 216 - Restrict SR COMPOSITE,
IMAGE and WAVEFORM to one reference. National Electrical Manufacturers Association
(NEMA); 2001. Available from: ftp://medical.nema.org/medical/dicom/final/cp216_ft.pdf
[Sup 23] DICOM Standards Committee. DICOM Supplement 23 - Structured Reporting Storage
SOP Classes. National Electrical Manufacturers Association (NEMA); 2000. Available from:
ftp://medical.nema.org/medical/dicom/final/sup23_ft.pdf
[TID 1400] National Electrical Manufacturers Association (NEMA). Digital Imaging and
Communications in Medicine (DICOM) Standard PS3.16 - Content Mapping Resource - TID
1400 Linear Measurement. Available from:
http://dicom.nema.org/medical/dicom/current/output/chtml/part16/chapter_A.html#sect_TID_140
0
[TID 1401] National Electrical Manufacturers Association (NEMA). Digital Imaging and
Communications in Medicine (DICOM) Standard PS3.16 - Content Mapping Resource - TID
1401 Area Measurement. Available from:
http://dicom.nema.org/medical/dicom/current/output/chtml/part16/chapter_A.html#sect_TID_140
1
[TID 1501] National Electrical Manufacturers Association (NEMA). Digital Imaging and
Communications in Medicine (DICOM) Standard PS3.16 - Content Mapping Resource - TID
1501 Measurement and Qualitative Evaluation Group. Available from:
http://dicom.nema.org/medical/dicom/current/output/chtml/part16/chapter_A.html#sect_TID_150
1
[DICOM CP 1852] DICOM Standards Committee. DICOM CP 1852 - Add code for bounding box
for SR. National Electrical Manufacturers Association (NEMA); 2019. Available from:
ftp://medical.nema.org/medical/dicom/final/cp1852_ft_addboundedby.pdf
[PS3.3 C.8.8.6.3] National Electrical Manufacturers Association (NEMA). Digital Imaging and
Communications in Medicine (DICOM) Standard PS3.3 - Information Object Definitions -
C.8.8.6.3 Representing Inner and Outer Contours on an Image. Rosslyn, VA: National Electrical
Manufacturers Association (NEMA). Available from:
http://dicom.nema.org/medical/dicom/current/output/chtml/part03/sect_C.8.8.6.3.html
[LIDC XML 2010] LIDC Team. Annotated XML File. 2010. Available from:
http://wiki.cancerimagingarchive.net/download/attachments/3539039/annotated_xml_file_Mar%
202010.rtf
[Fedorov 2020] Fedorov A, Hancock M, Clunie D, Brochhausen M, Bona J, Kirby J, et al.
DICOM re-encoding of volumetrically annotated Lung Imaging Database Consortium (LIDC)
nodules. Medical Physics. 2020;47(11):5953–65. Available from:
http://aapm.onlinelibrary.wiley.com/doi/abs/10.1002/mp.14445
[GeoJSON RFC] Gillies S, Butler H, Daly M, Doyle A, Schaub T. The GeoJSON Format. 2016
Aug . RFC 7946. Available from: http://tools.ietf.org/html/rfc7946
[DICOM CP 439] DICOM Standards Committee. DICOM CP 439 - Clarification of Structures
with Excluded Volumes in RT. National Electrical Manufacturers Association (NEMA); 2004.
Available from: ftp://medical.nema.org/medical/dicom/final/cp439_ft.pdf
[DICOM CP 2037] DICOM Standards Committee. DICOM CP 2037 - Inner and Outer Contours.
National Electrical Manufacturers Association (NEMA); 2020. Available from:
ftp://medical.nema.org/medical/dicom/final/cp2037_ft_InnerAndOuterContours.pdf
[Weerakkody] Weerakkody Y, Thuaimer A. Cobb angle. Radiopaedia. Available from:
http://radiopaedia.org/articles/cobb-angle
[ClearCanvas ShowAngles] ClearCanvas Workstation Clinical User’s Guide - Show Angles.
Available from:
https://documentation.clearcanvas.ca/Documentation/UsersGuide/Workstation/3_0/Clinical/inde
x.html?show_angles.htm
[Cornerstone.js CobbAngle] Cornerstone.js Cobb Angle Tool. Available from:
http://tools.cornerstonejs.org/examples/tools/cobb-angle.html
[Gaillard] Gaillard F. Acetabular angle - diagram. Radiopaedia. 2008. Available from:
http://radiopaedia.org/cases/acetabular-angle-diagram