Abstracting and Generalising theFoundational Model of Anatomy

Ontology

Eleni Mikroyannidi, Alan Rector, Robert Stevens

BioHealth Informatics GroupUniversity of Manchester

Outline

• Motivation: – The FMA is a highly detailed ontology of the human

anatomy• Consists of approximately 70,000 concepts• The use of the complete ontology is a time consuming procedure

– Why is the FMA so big?• How to make it smaller and more manageable• Without loss of information

• Methods & Evaluation Methodology– Abstraction– Expansion

• Results – Metrics • Conclusions

Motivation – One reason for size

• Symmetrical parts are represented exhaustively– e.g. Left_hand and Right_hand

• Three times as many entities as necessary– Left_hand, Right_hand and Hand

• Side benefit– Quality assurance

• Apparent omissions in restrictions and classes– e.g. Left_shoulder has fewer restrictions than

Right_shoulder

Methodology

• Abstraction– Abstract common characteristics of symmetrical

parts• From Hand, Left_hand, and Right_hand to just Hand• Plus the information that hands can be left or right

– the “Selector pattern”

• Expansion– Re-expand to original form

• Check that all information has been retained– And apparent omissions fixed

The aim of the Abstraction

• A smaller and more compact FMA without any loss of information– Consistent ontology design pattern (ODP) for symmetrical

structures• selector pattern to express symmetries

• Detect and report omissions in the FMA• Mechanism• Apply to a subset of the FMA with many symmetries

– The limbs• Used the OWL version of the FMA from Noy modified to be

more convenient– All partonomy relations moved from individuals to class

restrictions

Abstraction Algorithm

• User defines the symmetries and the ontology in the input• Creation of the Selector hierarchies• Detection and abstraction of the symmetrical entities

– A number of preconditions are checked • All the symmetrical classes are present• The symmetrical classes have equal number of restrictions • Lexically the label of the symmetrical class differs in one word from

the label of the parent class– Rejected candidates for the pattern are reported in log files

Abstraction Algorithm (1)

• The algorithm creates an additional restriction in the concepts referring to a symmetry– E.g. Hand has_laterality

some Laterality

Abstraction Algorithm (2)

• Common restrictions are moved in to the parent concept

• Symmetrical classes are removed

• Generalisation of the abstraction in order to consider other symmetries – E.g. Posterior and

Anterior, Medial and Lateral

Expansion Algorithm

• Detects classes with existential restrictions referring to the Selector– E.g. has_laterality some

Laterality

• Creates new symmetrical sibling classes– their label is based on

the parent class label• “Symmetry type +

parent class name”

Expansion Algorithm (1)

• Creates the existential restrictions of the symmetrical classes– Based on the

restrictions of the parent class

• Lexically modified and adapted

– E.g. regional_part some Left_thumb

Results - Metrics

• Abstraction metrics– FMA shrinks up to 57% maximum

• Expansion metrics– All the concepts are recreated in most cases– Some loss of restrictions when many symmetries are considered

• Omissions in the FMA• Extension of the algorithm to track reliably all the restrictions

Table 1. The number of classes of the (B.) abstracted and (C.) expanded FMA for each kind of symmetry (A.) that is added in the Symmetry hierarchy.

A. Type of symmetry that is added in each test

B. Class count of the abstracted FMA

C. Class count of the expanded FMA

Laterality selector 3149 6483

Metacarpal_bone_selector & Metatarsal_bone_selector

3032 6486

Anterior_Posterior_Selector

3029 6484

Middle_Distal_Proximal 2787 6482

Medial_Lateral_Selector 2782 6482

The number of classes of the initial FMA subset is 6483.

A. Type of symmetry that is added in each test

B. Subclass axioms count of the abstracted FMA

C. Subclass axioms count of the expanded FMA

Laterality selector 11076 19453

Metacarpal_bone_selector & Metatarsal_bone_selector

10628 18892

Anterior_Posterior_Selector 10617 18868

Middle_Distal_Proximal 9944 18430

Medial_Lateral_Selector 9913 18374

The number of subclass axioms of the initial FMA subset is 18477.

Table 2. The number of subclass axioms of the (B.) abstracted and (C.) expanded FMA for each kind of symmetry (A.) that is added in the Symmetry hierarchy.

Conclusions

• A more compact ontology as output• Same information of the FMA subset was

provided with fewer classes and axioms• No semantic errors were traced in both

abstracted and expanded ontology• The methodology was generalised to cover

different symmetries • It affords opportunities for quality assurance• Further development of the algorithm in order to

consider annotations of the symmetrical classes