Ontological support for Alzheimer’s disease aid to diagnostic
 
Authors:
Hanane Houmani, Centre de recherche Université Laval Robert-Giffard; Burt Crépeault; Simon Duchesne
 
Hypothesis:
The purpose of this project was to develop an ontological representation able to classify various types of knowledge (clinical, cognitive, genetic, and neuroimaging results) within the context of AD research: the Decision Support in Alzheimer’s Disease Ontology (DSADO).
 
Introduction:
Early detection of Alzheimer’s disease (AD) is critical for treatment success and constitutes a high priority research area. Post-mortem, pathologically-confirmed diagnostic accuracy of baseline clinical testing for AD averages 78% (22% error rate) (Weiner 2005) with insufficient diagnostic specificity (Dubois, Feldman et al. 2007). To increase the accuracy of this procedure, recent thinking on clinical diagnostic criteria for AD (Dubois, Feldman et al. 2007) stresses the importance of relying on a combination of core (clinical/cognitive) and supportive (neuroimaging/genetic/proteomic) assessment techniques (Dubois, Feldman et al. 2007). The combination of these techniques implies the integration of heterogeneous and complex knowledge.

Ontologies, as formal conceptualizations, constitute an essential step enabling interoperation in an heterogeneous and large domain knowledge. In fact, ontologies (Guarino 1995) enable sharing of information inside an area or between different areas in a consistent way. They further allow the researcher communities to agree on some concepts and their relationships (semantics), thus removing ambiguities related to a specific domain. Moreover, developing ontologies not only allows answers to standard queries, but also answers to conceptual queries in the structure of the domain that is not possible with classical techniques and methodologies (Lytras 2003).

Our domain can be defined as decision support in Alzheimer’s disease. However, there remains a need for an overall domain ontology covering the diagnostic procedure in AD, structured in such a way as to incorporate different and heterogeneous knowledge (clinical, cognitive, genetic, proteomic, and neuroimaging), and support the development of a reasoning engine that could provide decision support to the research communities in the diagnostic of AD.
 
Methods:
To develop this ontology, we used the On-To-Knowledge Methodology (Sure, Staab et al. 2002). The choice of the On-To-Knowledge Methodology was motivated by two reasons. First, it conforms to the IEEE 1074-1995 standard (IEEE 1996), which ultimately helps achieve quality ontologies. Second, the methodology is appropriate as the developed ontology is part of an ongoing project to design a reasoning based system.

We further constrained our design by following these principles:

  • whenever possible, reuse publicly available ontologies;
  • wherever public ontologies did not exist, create novel entities based on established terminologies (e.g., UMLS [Lindberg, Humphreys et al. 1993])
  • ensure that the ontology was compatible with current, established standards for medical informatics (e.g., DICOM [2001])
 
Results:
The Decision Support in Alzheimer’s Disease Ontology is described in Figure 1.

Figure 1

Figure 1 - Decision Support in Alzheimer’s Disease Ontology. Ontologies in red are public ontologies that are being reused, while blue ontologies will need to be developed.

The SUBJECT is the entry point to the ontology. A SUBJECT is an individual with associated invariant demographics information covered in the ontology, such as age and gender. Each SUBJECT has:

  • RELATIVES, that could be themselves subjects. This is used to represent the familial history of the subject. We reuse the Family History Health ontology (Peace and Brennan 2007);
  • BIOFEATURES, which represent biological characteristics. The BIOFEATURES ontology is composed of sub-ontologies such as ANATOMY (based on the Foundational Model of Anatomy – NeuroNames (Hole and Srinivasan 2003); PHYSIOLOGY; METABOLISM; NEUROPSYCHOLOGY; GENETICS; and PROTEOMICS;
  • STUDIES, which investigate a DIAGNOSTIC for the SUBJECT during various SESSION.

The DIAGNOSTICS ontology covers ante (e.g., none; undefined; possible; probable) and post-mortem (definite) diagnostics. This ontology expresses the possible diagnostics investigated by the STUDY, evolving through time in various SESSIONS.

For each DIAGNOSTIC is associated a HUMAN DISEASE. This particular ontology is reused from Maja et al. (Maja and Elizabeth 2005). It includes sub-ontologies such as TYPES, PHENOTYPES, CAUSES, and TREATMENTS.

Each STUDY has associated ETHICS that represents the extent and type of consent, and instructions as to the use of personal data. Each STUDY concerns one or more temporal SESSIONS. Each SESSION in the diagnostic process has:

  • Associated PHENOTYPES, describing the symptoms experienced by the SUBJECT at this point in time. Note that we use the PHENOTYPES sub-ontology of HUMAN DISEASES;
  • An ENVIRONMENT, covering time-dependent epidemiological data such as medication history, smoking, place of residence;
  • REPRESENTATIONS, as embodiments of medical tests such as neuroimaging, genetics testing, blood or cerebro-spinal fluid sampling. A MEDICAL DEVICE performs each REPRESENTATION. We base both of these ontologies on the DICOM standard.

The REPRESENTATION is processed via a given METHOD, producing a MARKER that serves as a proxy of a specific BIOFEATURE. The MARKER ontology covers core and supportive features, as proposed in Dubois et al. (REF). For example, manually obtained volume (MARKER) of the hippocampus is a specific proxy for true hippocampal volume (BIOFEATURE), and can be obtained via tracing (METHOD) on a T1-weighted image (REPRESENTATION) from a magnetic resonance scanner (MEDICAL DEVICE).

Each METHOD can be described by REFERENCES; the latter represents the ontology of references, journals and authors, as described in details by the Semantics Web Application in Neuromedicine (SWAN) ontology (Paolo, Elizabeth et al. 2008). A METHOD is subjected to a Verification, Validation and Evaluation process that will be captured by the VVE ontology (Jannin 2006). Finally, each MARKER has an associated COST.

Notice that REPRESENTATIONs and MARKERs are multi-modal, in that they can be clinical, cognitive, proteomic, genetic or neuroimaging results. Indeed, these concepts generalize the fact that we have knowledge that comes from different biomedical domain.
 
Discussion:
In spite of the large variety of ontologies in the biomedical field, to our knowledge no ontology has modeled the diagnostic of Alzheimer’s disease, specifically by incorporating heterogeneous and complex knowledge. The DSADO model could easily be extended to include the domain of interest of diagnostic in other neurodegenerative diseases, such as other dementias (e.g., fronto-temporal, vascular, Lewy bodies).

We have attempted to reuse publicly available ontologies (e.g., SWAN, HUMAN DISEASES). Where needed, we elected to construct our own, which will be returned to the public domain as part of our project.

 
Conclusion:
We propose in this paper an Ontology-based Support for Alzheimer’ Disease Aid to Diagnostic. This ontology allows us to incorporate complex and heterogeneous knowledge coming from different bio-medical fields, such as clinical, cognitive, genetic and neuroimaging testing. In addition, it allows us to model the diagnostic process in AD. Testing and validation will follow the ontological development. Our first task will be to use real life samples, such as the Alzheimer’s Disease Neuroimaging Initiative dataset (Mueller, Weiner et al. 2005), to develop knowledge-based reasoning that may achieve a more accurate assessment of AD than either of these knowledge sources can do alone.
 
References:
“DICOM reference guide.” Health Devices. 2001;30(1-2):5-30.

Dubois B, Feldman HH, et al. “Research criteria for the diagnosis of Alzheimer’s disease: revising the NINCDS-ADRDA criteria.” Lancet Neurol. 2007;6(8):734-46.

Guarino N. “Formal ontology, conceptual analysis and knowledge representation.” Int. J. Hum.-Comput. Stud. 1995;43(5-6):625-640.

Hole WT, Srinivasan S. “Adding NeuroNames to the UMLS Metathesaurus.” Neuroinformatics. 2003;1(1):61-3.

IEEE, S. IEEE standard for developing software life cycle processes S.E. S. C. of the IEEE Computer Society, IEEE. 1996.

Jannin PG, Maurer CR Jr. “Model for defining and reporting reference-based validation protocols in medical image processing.” International Journal of Computer Assisted Radiology and Surgery. 2006;1(2).

Lindberg DA, Humphreys BL, et al. “The Unified Medical Language System.” Methods Inf Med. 1993;32(4):281-91.

Lytras MDP, Korfiatis AN. An Ontology-Oriented approach on E-learning: Integrating Semantics for Adaptive E-learning Systems. ECIS. Naples, Italy. 2003.

Maja H, Elizabeth C. Ontology-Based Support for Human Disease Study. Proceedings of the Proceedings of the 38th Annual Hawaii International Conference on System Sciences (HICSS’05) - Track 6 - Volume 06, IEEE Computer Society. 2005.

Mueller SG, Weiner SG, et al. “Ways toward an early diagnosis in Alzheimer’s disease: The Alzheimer’s Disease Neuroimaging Initiative (ADNI).” Alzheimers Dement. 2005;1(1):55-66.

Paolo C, Elizabeth W, et al. “The SWAN biomedical discourse ontology.” J. of Biomedical Informatics. 2008;41(5):739-751.

Peace J, Brennan PF. “Ontological representation of family and family history.” AMIA Annu Symp Proc. 2007;1072.

Sure Y, Staab S, et al. “Methodology for development and employment of ontology based knowledge management applications.” SIGMOD Rec. 2002;31(4):18-23.

Weiner M, et al. The Use of MRI and PET for Clinical Diagnosis of Dementia and Investigation of Cognitive Impairment: A Consensus Report. , Alzheimer’s Association. 2005.