Anatomist Structural Abstraction
Cornelius Rosse, M.D., D.Sc., José L. V. Mejino, M.D., Linda G. Shapiro, Ph.D.,
James F. Brinkley, M.D., Ph.D.
Structural Informatics Group, Departments of Biological Structure and Computer Science, University of Washington, Seattle, WA
The Visible Human data sets have stimulated a great deal of activity in the graphical representation of anatomy. A major challenge is to enhance this resource of image-based information with knowledge of its own structure. There is a need for a symbolic model of the structural organization of the human body, which could invest with meaning the graphical information extractable from the clusters of voxels and their geometric coordinates that make up the Visible Human data sets. The objective of this communication is to examine the elements of structural information such a symbolic model should encompass, and to assess the extent to which the Anatomical Structural Abstraction (ASA) of the Digital Anatomist Foundational Model of Anatomy (Fm)1 meets this objective.
Elements of Structural Information.
In a biological or anatomical context, the term structure
is associated with two distinct concepts (meanings): 1. a material object
generated as a result of coordinated gene expression, which necessarily consists
of parts (e.g., hemoglobin molecule, cell, heart, human body); and 2. the manner
of organization or interrelation of the parts that constitute a structure
specified by the first definition (i.e.,
the structure of a structure). Both definitions emphasize the critical need for
declaring the principles according to which units of organization can be defined
in order to be able to state what is ‘whole’ and what is ‘part’.
Specifying the manner in which parts interrelate must satisfy two requirements:
1. to determine the kinds of parts of which various structures may be
constituted; and 2. to state the manner of spatial organization of parts by
describing their boundaries, continuities and attachments, as well as their
location, orientation and spatial adjacencies in terms of qualitative
coordinates (in addition to the quantitative geometric coordinates, which are
embedded in the Visible Human data sets).
The Anatomical Structural Abstraction.
is one of the four components of the Digital Anatomist Foundational Model of
= (Ao, ASA, ATA, Mk)
is an abstraction that, in accord with declared principles, describes the
physical organization of the material objects and 3D spaces that constitute an
idealized human body1. The current model is limited to the static
state and excludes anatomical structures smaller than the cell.
Its backbone is the Anatomy ontology (Ao),
which assigns anatomical entities to classes according to defining attributes
they share with one another and by which they may be distinguished from one
the subject of this communication, is described after other Fm
components. ATA, the Anatomical
Transformation Abstraction describes time-dependent morphological
transformations of anatomical structures during the human life cycle. Mk,
Metaknowledge, comprises the
principles, definitions and rules according to which relationships are
represented in the other three components of Fm.
captures the information that is sufficient and necessary for describing the
structure of any physical object or space that constitutes the body, as well as
that of the entire human body itself. The ASA of the entire body may be conceived as a composite of all ASAs
of anatomical structures and anatomical spaces that are represented in Ao.
The Anatomical Structural Abstraction is distinct from other approaches that
have been proposed for the symbolic description of anatomical spatial
relationships, in that it is not limited to object recognition in medical
images, it generalizes to all parts of the body and
it accommodates all relationships that are necessary for
describing the 3D structure of the body. The ASA consists of several
= (So, Pn, Bn, SAn)
So = Spatial object
Pn = Part-of network
Bn = Boundary network
SAn = Spatial association network
provides an additional axis for classifying the thousands of anatomical concepts
in Ao according to their spatial dimensions and shape, and thereby
systematizes the description of their spatial relationships. For instance,
through the rules entered in Mk,
part-whole relationships in Pn
are restricted to spatial objects of the same dimension. For example, the
relationship ‘Right atrium’ -has
parts- ‘Cavity of right atrium’,
‘Wall of right atrium’ is sanctioned because in So
all these concepts are classified as Volume (3D object). On the other hand, ‘Surface
of right atrium’ cannot be modeled as part of ‘Right atrium’, because it is a 2D object. The correct relationship
is specified by Bn: ‘Right atrium’
-bounded by- ‘Surface of right atrium’,
because the latter is classified as a spatial object of one lower dimension than
the atrium. Pn also incorporates an ontology of part-whole relationships,
which are sanctioned according to the assignment of a concept to classes of Ao
For instance the relationship -has lobe- is sanctioned if the ‘whole’ is
classified as a “Parenchymatous organ’ in Ao
and a ‘Cone’ ‘Semicone’ or ‘Polyhedron’ in So.
Spatial Association Network (SAn) itself encompasses
subnetworks of various relationships:
SAn = (Ctn, Atn, Ln,
On, Ajn) (3)
= Continuity network
Atn = Attachment network
Ln = Location network
On = Orientation
Ajn = Adjacency network
The latter three subnets of the SAn make use of
traditional anatomical descriptors of orientation and location (anterior,
posterior, etc.) as qualitative coordinates in terms of the shape of the spatial
object that is being described.
We began to represent the Fm as an extension and enhancement of the UMLS semantic network.
A semantic net, however, is not sufficiently expressive for modeling the
multiple relationships that constitute the ASA. Therefore, we are in
the process of migrating the model to Protégé, a frame-based system designed
for accommodating multiple relationships, supporting inheritance and correlating
several ontologies3. The Ao, So and Pn
components of Fm, along with relevant elements of Mk have been implemented
for macroscopic anatomy. Work is in progress on the structural networks of the ASA.
The Visible Human geometric data sets serve
for verifying spatial relationships we model and also for validating the ASA
schemes that we develop. This work would be greatly facilitated by the
availability of 3D graphical models for all anatomical structures that can be
segmented from the volumetric data.
It is our contention that the Digital Anatomist
Foundational Model as a whole, and its Anatomical Structural Abstraction in
particular, will furnish the formal representation of knowledge that will
provide for the intelligent navigation of the Visible Human data sets. Through
appropriately designed interfaces, the Fm
will be instrumental in revealing the structure of the human body not only to
experts but also to any user who has a need for interacting with the Visible
Supported by National Library of Medicine contract, LM
83510 and grants LM 06822 and LM 06316.
C, Shapiro LG, and Brinkley
JF. The Digital Anatomist Foundational Model: principles for
defining and structuring its concept domain.
J Am Med Inform Assoc Proc AMIA’98 Annual Symposium 1998;820-824.
C, Mejino JL, Modayur BR, Jakobovits R,
Hinshaw KP, Brinkley JF. Motivation and organizational principles for
anatomical knowledge representation: the Digital Anatomist Symbolic Knowledge
Base. J Am Med Inform Assoc 1998;5:17-40.
MA, Gennari JH, Eriksson H, Tu SW, Puerta AR. PROTÉGÉ II:computer support for
development of intelligent systems from libraries of components. MEDINFO95, The
eighth World Congress of Medical Informatics, Vancouver, B.C. Canada,