- As biomedical ontologies grow in size and complexity it is crucial to develop methods for detecting inconsistencies within ontologies. The Foundational Model of Anatomy (FMA) ontology represents knowledge of human anatomy, with structural organization provided by class and part relationships. Using a manual audit, I identify types of inconsistencies arising from class and regional part relationships for regions of the body and the parts of organs. Inconsistencies arise from both explicitly declared relationships and relationships that are implied by the lexical constructs of class names. The purpose of this work is to propose methods of structural organization and lexical consistency that will make the FMA more compatible with computational auditing and increase its usability.
I. INTRODUCTION
As biomedical ontologies grow in size, specificity and complexity, maintaining internal consistency of the representation becomes increasingly difficult. Because ontologies can contain tens of thousands of classes, computational methods for detecting inconsistencies are necessary for quality assurance efforts. One step toward designing computational auditing methods is to identify patterns within the content that are useful for identifying inconsistencies in modeling or possible errors.
This paper examines part and class relationships within the Foundational Model of Anatomy (FMA) ontology—both those explicitly declared and those implied by the lexical constructs of class names. The purpose of this work is to identify types of inconsistencies and to propose methods of structural organization and lexical consistency that will make the FMA more compatible with computational auditing and increase its usability.
The FMA represents knowledge of human anatomy using
a series of organizing units at different levels of granularity.
These units include “Cardinal body part”, “Organ system”,
and “Organ” (Fig. 1) [
Fig. 1. Overview of part relationships using high-level classes within the FMA. Classes representing organizing units are in the blue bar. Typical regional_part and constitutional_part relationships among the subclasses are indicated with solid black arrows. SubClass_of relationships are indicated with gray dashed arrows.
Each of the cardinal body parts (“Head”, “Body proper”, left and right “Upper limb”, and left and right “Lower limb”) are divided into a hierarchy of smaller regions using the regional_part relation. These divisions of the body are crucial in helping users understand and navigate the FMA and to provide classes that can be lexically modified to create other classes.
As demonstrated in Fig. 2, there is an opportunity to tightly align the regional part hierarchy for cardinal body parts to the class hierarchy. Much of the structure for coordinating the two hierarchies is already in place, but some regions of the body are classified as types of anatomical clusters instead of subdivisions of the body. This obscures the relationship of those classes to other regions of the body.
There are two benefits to formalizing this organizational structure. First, it provides a place to classify all divisions of a particular region—even if a class represents an alternative partition which is not reflected in the regional_part hierarchy. This is important for classes such as “Trunk”, which have more then one partition scheme. Second, users could more easily answer the question of what partitions are provided by the FMA for a particular region of the body by navigating the class hierarchy beneath “Subdivision of cardinal body part”. For example, classes such as “Right side of head” and “Left side of head” do not currently have any relationships to other classes. But because they are subclasses of “Subdivision of head”, users can locate these terms to use in annotation.
As shown in Fig. 3, some class names consist of a modifier (such as “Surface of…”) and a body region or high-level class for types of body regions. Consistency in referring to body regions would aid auditing and ontology users. In cases where another anatomical term is traditionally used, the term reflecting a consistent construct could be added as a synonym.
Another source of inconsistency arises from the lexical implications of class names and their positions within the class hierarchy. As shown in Fig. 4, “Integument of chin” is a subclass of “Integument of subdivision of mouth.” From this a user would infer that “Chin” is a “Subdivision of mouth”, which is confirmed by the subclass relationship for “Chin” found in the FMA. However, as detailed in Fig. 4, contradictions can easily arise. For example, “Set of jaws” is a subclass of “Set of subdivisons of head”, but “Jaw” is a subclass of “Organ cluster.” Correcting these inconsistencies depends upon first establishing a comprehensive regional_part hierarchy for subdivisions of cardinal body parts.
The FMA has a complex representation of organ parts. Parts are classified as either an “Organ component” (bounded predominately by bona fide boundaries) or an “Organ region” (defined by fiat boundaries). Each organ region is further classified as an “Organ segment” (with anchored fiat boundaries) or an “Organ zone” (with floating fiat boundaries).
As shown in Fig. 5, inconsistencies arise when the lexical implications of class names do not match the class hierarchy. For example, “Region of crown of tooth” is a subclass of “Region of organ component”. However, “Crown of tooth” is classified as an “Organ segment”.
It is unclear if the complexity of representation for organ parts adds value to the FMA. But if retained, then auditing methods should be developed to ensure consistency.
Several types of organs have a tree structure, including arterial tree organs and venous tree organs. The regional parts of these trees can be described as a trunk plus either branches or tributaries. The FMA has two specific types of regional_part relations (branch and tributary) to relate branch and tributary parts to the tree. However these relations are applied inconsistently—duplicating the regional_part relationship, substituting for the general relationship, or not used at all. This inconsistency impacts efforts to access knowledge about tree organs and determine the completeness of representations.
The terms “branch” and “tributary” have specific anatomical meanings in the FMA. But for other terms such as “subdivision”, “region”, “segment”, and “subsegment” it is unclear whether these are synonyms or they carry specific meaning when used in class names.
The term “portion” should be reserved for subclasses of “Portion of tissue” and “Portion of body substance”, but has been applied to several regions of organs (such as “Intrapulmonary portion of pulmonary artery”).
I have documented a variety of ways that inconsistencies in the representation of regional parts occur within the FMA. The first step to addressing these inconsistencies is to establish a robust regional part representation of the cardinal body parts.
Previous work to audit the FMA has used symmetric terms
(for example, “left” and “right”, “superior” and “inferior”) [
I thank José L. V. Mejino Jr. for helpful discussions. This work was supported in part by funding from the International Health Terminology Standards Development Organization (IHTSDO) to the Structural Informatics Group.