1 ios press ontology alignment for wearable devices and ... ?· j. hodges et al. / ontology...

Download 1 IOS Press Ontology alignment for wearable devices and ... ?· J. Hodges et al. / Ontology alignment…

Post on 05-Jun-2018




0 download

Embed Size (px)


  • Undefined 1 (2014) 15 1IOS Press

    Ontology alignment for wearable devices andbioinformatics in professional health careEditor(s): Name Surname, University, CountrySolicited review(s): Name Surname, University, CountryOpen review(s): Name Surname, University, Country

    Jack Hodges a,, Mareike Kritzler a and Florian Michahelles a and Stefan Lueder aand Erik Wilde aa NEC WOT-US, Siemens Corporate Technology, 2087 Addison, Berkeley, CA 94074, USAE-mail: {jack.hodges.ext, mareike.kritzler, florian.michahelles, stefan.lueder, erik.wilde.ext}@siemens.com

    Abstract. Web Ontology Language (OWL) based models and triple stores hold great potential for access to structured infor-mation. Not only are OWL-based ontologies extremely versatile and extendable, but triple stores are robust against changes toontologies and data. The biomedical field illustrates this value insomuch as it employs vast amounts of information distributedacross different models and repositories. This paper presents a case study that sought to demonstrate the real-world value oflinking disease, symptom, and anatomical models with wearable devices and physical property models and repositories. Integrat-ing these models is both necessary and problematic; necessary to provide undifferentiated access to health care professionals,problematic because although the biomedical ontologies and repositories exist, they arent semantically aligned and their designsmake alignment difficult. This case study demonstrated that manually linking multiple biomedically-related models can producea useful tool. It also demonstrated specific issues with aligning curated ontologies, specifically the need for compatible ontologydesign methodologies to ease the alignment. Although this study used manual ontology mapping, it is believed that systems canbe developed that can work in tandem with subject matter experts to reduce mapping effort to verification and validity checking.

    Keywords: ontology alignment, semantic bridge, professional health care, wearable devices, physical properties

    1. Introduction

    Consider a physician that has a patient diagnosedwith Type II Diabetes. How might this physician lo-cate wearable devices that could aid in tracking the pa-tients condition? Currently they would have to enu-merate the symptoms associated with Diabetes, asso-ciate those with the kinds of measurements neededto qualify/quantify the symptom, and then search forwearable devices capable of measuring that quantity.Using a semantic integration it would be possible to doall this in a single application.

    Achieving the promise of semantic technology in-volves a mixture of upper and domain ontologies, se-mantic repositories, and integration (linking and map-

    *Corresponding author. E-mail: editorial@iospress.nl.

    ping) across ontologies. For ontologies such as QUDT(quantities, units, dimensions, and datatypes) [4] therelated ontologies and repositories are fully integrated,meaning that they act as a single model. The vast ma-jority of curated ontologies/repositories, however, re-main stand alone. For example, the biomedical fieldhas produced many ontologies; for diseases, biochem-istry, symptoms, anatomy, pathology, etc. For the mostpart these ontologies are not integrated and thus theinformation they model cannot be shared across witheach other.

    This case study sought to demonstrate the value ofsemantic integration by showing how medical profes-sionals could benefit by having integrated access tobiomedical models/repositories that also integrate withwearable devices and the properties they measure. Inthe process of integrating the information sources thatwould allow the above example to be realized, both

    0000-0000/14/$00.00 c 2014 IOS Press and the authors. All rights reserved

  • 2 J. Hodges et al. / Ontology alignment for wearable devices and bioinformatics in professional health care

    ontology and data mapping issues were encountered.This paper describes the mechanisms used to achievethis integration, manually, using a subset of five on-tologies. The conclusion borne out by this study sug-gests that ontology mapping can be made easier withthe use of computational techniques but still requiressubject matter experts for validation and verification.

    2. Goals

    There were 4 primary goals in this study, all associ-ated with the desire to show that integration of struc-tured information sources would be useful for medicalprofessionals:

    Demonstrate the integration of standardized andexternally-curated models

    Demonstrate that linked biomedical models pro-vide value through single-point access

    Demonstrate integration through the use of quan-tity/unit/value models

    Demonstrate the ability to associate wearable de-vice models with biomedical models

    The first goal saves us from having to take the timeand energy to create models from scratch, but ulti-mately causes two problems inherent in any integra-tion project; (1) mismatched design or implementa-tion approaches, and (2) incomplete data. The remain-ing goals are illustrative of the functionality this studyhoped to demonstrate; namely to bring together a setof ontologies that would connect wearable sensors andmedical professionals.

    The information types this study sought to connectare shown in Figure 1:

    Fig. 1. Targeted integration of 5 information types

    Figure 1 shows five information types. First, itwas desirable to integrate biomedical information thatwould be useful in helping medical professionals eval-uate/select wearable devices for patients: (1) humandiseases, (2) disease symptoms, and (3) human gross

    anatomy. For the physician, being able to diagnose apatient, and relate the associated disease or symptomswith associated anatomy, would allow multiple accesspoints to integrate with wearable devices.

    Second, symptoms are associated with physicalproperties (e.g., cardiac disease and blood pressure),so integrating symptom models with properties that awearable device might measure is necessary.

    Finally, to integrate the wearable device to thebiomedical models, both the anatomical parts wherethe device is worn and the properties that it measuresare needed.

    2.1. Choice of Ontologies

    The choice of ontologies used in this study wasbased on the goal of demonstrating an integratedvalue. Three ontologies from the Open Biological andBiomedical Ontologies (OBO) Foundry ontologieswere selected due to the content they modeled as wellas the fact that they had unique identifiers that allowedthem to be cross referenced. Disease information wasrepresented using the OBO Disease Ontology (DOID)[12], symptom information was represented using theOBO Symptoms Ontology (SYMP) [13], and anatom-ical information was represented using the OBO Foun-dation Model of Anatomy (FMA) ontology [9]1 wereused.

    The QUDT[4] models were used because they rep-resent an integrated approach to quantities, units,dimensions, and datatypes, but other models existthat might have been used, such as the Model Li-brary for Quantities, Units, Dimensions, and Values(QUDV) [10], the Library for Quantity Kinds andUnits (QU)[15]. QUDT was deemed to be more com-prehensive set of models, and had representations forbiomedical quantities needed to model wearable de-vices.

    The Semantic Sensor Network (SSN) [7] ontol-ogy is an emerging standard for device modeling andseemed a good integration point for the project.

    3. Approach

    From a functional point of view the study would besuccessful if all 5 semantic models can be integratedinto effectively a single model and searched from mul-

    1All available through the Open Biological and Biomedical On-tologies Foundry


  • J. Hodges et al. / Ontology alignment for wearable devices and bioinformatics in professional health care 3

    tiple entry points in a single interface. The integrationrequired minimally 5 property alignments (and associ-ated mappings) between the ontologies:

    Diseases (DOID) symptoms (SYMP) Symptoms (SYMP) anatomy (FMA) Symptoms (SYMP) properties (QUDT) Wearable devices (SSN) quantities (QUDT) Wearable devices (SSN) anatomy (FMA)

    These alignments are based on functional dependen-cies between the models. For example, there wouldbe no need to align wearable devices with diseasesor symptoms because there is no functional relation-ship between them. On the other hand, an alignmentcould be modeled between diseases and anatomy, butsince there is a direct relationship between disease andsymptom, and between symptom and anatomy, the re-lationship between disease and anatomy can be in-ferred and need not be explicitly modeled.

    The ontological integrations used to support seman-tic search in this study are shown in Figure 2:

    Fig. 2. Integration of 5 ontologies and sample graph traversal

    Figure 2 illustrates the ontological alignments usedin this study along with some values for a particularsearch task (in this case, Type II Diabetes). The largedotted line shows one path (disease symptom property device) a search might take through thesemodels to relate Type II Diabetes to a wearable devicecapable of measuring heart rate/blood pressure for thesymptom abnormal weight gain. Solid lines betweenthe repositories represent the ontological alignments.The data sets are depicted as repositories to illustratethe integrative/federated nature of the study. Squareboxes represent items in the repositories along the de-picted inference path.

    3.1. Alignment Problems

    There are three aspects to the model integrationproblem encountered in this study. First, the semantic

    structure aligning the models must exist. That is, theremust be classes and class properties that support thealignment. Second, the data allowing for an integrationmust exist. F