Knowledge Management in Medicine

A Framework to Organize, Browse and Retrieve Medical Data

Marios Pitikakis

and Imon Banerjee

Softeco Sismat S.r.l., Via De Marini 1, Genova, Italy

CNR IMATI-Genova, Via De Marini 6, Genova, Italy

Keywords: Knowledge Management, Bio-medical Ontology, Medical Imaging, Computer-Aided Diagnosis.

Abstract: This paper outlines a knowledge-based approach to construct semantically enriched components that can be

integrated to form an environment for assisting and supporting medical professionals. We aim at capturing

and utilizing invaluable expert knowledge, formalized within ontologies, to improve different kinds of

medical diagnosis, monitoring and treatment. Our work is focused on supporting the knowledge

management task, targeting the early stage diagnosis of musculoskeletal diseases of the human knee

articulation, but is general enough to support similar knowledge management tasks for a wide range of

clinical decision-making, research, teaching and learning activities.

1 INTRODUCTION

Knowledge is a very important resource for

preserving valuable information, solving problems

and creating core competences. Managing this

knowledge has become an important research issue

and a wide range of technologies for both academic

and real world applications have been developed.

In the medicine and health care domain, there are

many individual applications and tools that rarely

share common semantics. Queries related to medical

data (images, clinical records, treatments plans etc.)

are not arbitrary; they are based on specific

semantics of anatomy, physiology and diseases.

Medical databases have a wealth of digital

resources and formalized knowledge can help to

organize them in an efficient way in order to support

searching, browsing and retrieval tasks, as well as to

assist in the diagnosis and follow-up practices.

Ontologies can provide the essential glue to ensure

semantic consistency of data and knowledge sharing

by the different actors in complex medical scenarios.

In this work, we focus on developing an

ontology-based knowledge management framework

and a set of tools and services for computing

different kinds of diagnostic measurements which

can support the sharing, access, retrieval and

integration of various pieces of information related

to Musculoskeletal Diseases (MSD) and other

disorders of the human knee region. To this end, our

research activities take into account different scales

(i.e. molecular, cellular, tissue, organ and behavior

scales) and modalities in a Computer-Aided

Diagnosis (CAD) context. Combining ontologies

with CAD systems could improve the segmentation

and analysis processes as well as the follow-up and

treatment by applying generic knowledge to highly

patient specific data (Catalano et. al., 2012). On the

other hand, diagnostic measurements (e.g. cartilage

thickness map, bone/organ volume) from standalone

applications can create meaningful semantic

annotations/associations and contribute to the

knowledge formulation.

This paper aims at presenting a knowledge-based

approach targeted at constructing semantically

enriched components that can be integrated to form

an environment that can assist and support medical

professionals. Section 2 contains the motivation

behind our efforts and reviews related work. Section

3 shortly describes the conceptualization of the

domain and the purpose of the knowledge base.

Section 4 gives an overview of some preliminary

results in the form of standalone and web-based

applications. Finally, in Section 5 we present our

concluding remarks, the proposed future work, and

the envisaged research road-mapping.

374

Pitikakis M. and Banerjee I..

Knowledge Management in Medicine - A Framework to Organize, Browse and Retrieve Medical Data.

DOI: 10.5220/0004867603740380

In Proceedings of the International Conference on Health Informatics (HEALTHINF-2014), pages 374-380

ISBN: 978-989-758-010-9

 2014 SCITEPRESS (Science and Technology Publications, Lda.)

2 BACKGROUND AND RELATED

WORK

Computer-Aided Diagnosis (CAD) is one of the

major research subjects in medical informatics and

diagnostic radiology. CAD is a well established

concept where physicians use the computer output in

a complementary way to support their final

diagnosis (Doi, 2007).

The need to improve the quality of health care

has led to a strong demand for CAD systems that

can provide accurate, repeatable and objective

feature measurements which can be used by

physicians for diagnostic and follow-up purposes.

Current CAD approaches concentrate on algorithmic

improvements, and mostly fail to include the

implicit knowledge of medical applications.

Modern diagnosis depends on the patient’s health

condition acquired by different modalities. Most of

the times it becomes difficult to integrate all these

acquired data to obtain the final diagnosis, resulting

in a strong manual effort by experts of different

domains to combine and interpret these data.

Therefore, one of the biggest challenges is to design

a common platform for searching, browsing,

accessing and combining these data (preferably in an

automatic/semi-automatic way) for better diagnosis

and follow-up results. This kind of platform could

also offer the prospect of integration, or at least

interconnection, among doctors and medical

professionals and enable them to navigate more

easily through the data and directly gather all the

relevant information available.

Ontologies up to now were mainly used to

provide a common vocabulary in different domains.

Especially bio-ontologies are quite popular for

providing taxonomies and supporting different

knowledge management tasks. The use of ontologies

in medicine started with focus on the representation

and (re-)organization of medical terminologies, for

example FMA (Foundation Model of Anatomy)

(Rosse & Mejino, 2003), ICD (International

Classification Diseases, online), SNOMED

(Systematized Nomenclature of Medicine)

(Spackman, 2000) etc. These reference ontologies

are used to provide a common vocabulary between

the medical experts for the establishment of a shared

understanding of concepts used.

In addition, the advancement of semantic web

technologies contributed to the widespread usage of

ontologies and made possible to extract structural,

functional and morphological information from

heterogeneous medical data residing in different bio-

medical ontologies. This fact can potentially support

computational frameworks for clinical decisions e.g.

OntoQuest (Chen et. al., 2006), or study/analyze the

human anatomy and the functional behavior of

organs in a more interactive way e.g.

MyCorporisFabrica (MyCF) (Palombi et. al., 2009).

Furthermore, linking clinical knowledge with the

geometry extracted from the patient record is likely

to open new pathways for clinical analysis.

Extracting anatomically and functionally significant

regions from medical imagery is another challenging

and essential task. In the process of image

segmentation, it is highly beneficial to attach

semantic information to the segmented parts,

addressing not only standardization, but also

machine-readability (due to the formalized

representation), advanced browsing and searching.

Some efforts on semantic tagging can be seen in

the area of generic human body modelling for

teaching and training purposes, as well as patient

specific modelling for studying the patient condition.

The Zygote Body browser (Zygote browser, online),

Voxel man (Voxel-man, online) and BioDigital

Human (BioDigital Human, online) represent some

of the recent work on generic human body

modelling combined with semantic knowledge,

while 3D anatomical human (Magnenat-Thalmann

et. al., 2007) and MyCF browser (MyCF browser,

online) focus on patient specific models.

In medicine, this knowledge is useful to drive

automated analysis to support diagnosis, therapy

planning, surgery and legal medicine. Only a few

initiatives have taken on the use of geometric data

derived from acquired images and canonical

anatomic knowledge e.g. the Virtual Soldier project

(Virtual soldier project, online) by the U.S. Defence

Advanced Research Projects agency. The use of

semantic technologies to the creation of expert

systems applied to the medical diagnostic process is

presented in (Rodríguez-González et. al., 2012),

where a knowledge base containing findings (signs

and symptoms) and diagnostic tests was developed.

Formalizing and retrieving images and 3D data

from different medical acquisition devices is not a

trivial task, since their characterization depends on

morphological attributes as well as on other

semantic attributes. In this context, one of the main

results the AIM@SHAPE Network of Excellence

(AIM@SHAPE, 2006, online) was the formalization

and sharing of knowledge related to 3D digital

shapes and their applications. The scientific

community involved in AIM@SHAPE brought a

significant contribution to the development of

ontologies for 3D applications by proposing a

conceptualization of a shape, meant as a

KnowledgeManagementinMedicine-AFrameworktoOrganize,BrowseandRetrieveMedicalData

375

generalization of a dataset, in terms of geometrical,

structural and semantic aspects, complemented by

the knowledge related to the application domain in

which the shape is used. This resulted in the design

of two kinds of ontologies, namely the Common

Ontologies (for both Shapes and Tools) and three

Domain Ontologies. The Common Shape Ontology

(CSO) (Vasilakis et. al., 2010) was conceived to

capture knowledge related to the set of geometric,

structural and topological data that define a shape.

The Digital Shape Workbench (DSW) (Pitikakis

et. al., 2012) was one of the main results of

AIM@SHAPE, which was later upgraded to the

Virtual Visualization Services (VVS) of

VISIONAIR (Attene et. al., 2013). VVS is a

framework based on Semantic Web technologies for

managing, storing, reasoning and searching the

semantic content. It integrates resources and

knowledge, providing functionalities for inserting

resources, managing the resource metadata and

related ontologies, advanced searching, browsing

and downloading of resources.

3 SEARCHING AND BROWSING

THE KNOWLEDGE BASE

In a complex medical scenario where multiple

agents co-operate in order to allow continuity of

care, formalized knowledge can help to organize the

different resources e.g. acquired data, exams,

anatomical information, patient history and

links/relations between them.

Our work focuses on MSD and related disorders

of the human knee articulation. The main motivation

of this work is to develop a knowledge management

and ontology framework, which can facilitate the

sharing, access to, refinement and integration of

various pieces of information pertaining to the MSD

domain. A knowledge-driven framework provides

access and search functionalities to concepts,

patient-related data, and information related to

musculoskeletal pathologies, which are properly

addressed via a shared conceptualization.

For the formalization of the MSD domain, our

obvious choice was to utilize ontologies. We did not

start the ontology design, and the corresponding

conceptualization, from scratch, but we capitalized

on what other initiatives already built concerning

biomedical aspects of the domain. After analyzing

existing work and deciding what can be reused, we

planned our design process that included the

integration of existing ontologies or parts of them.

We choose the versatility of the middle-out approach

for ontology design, since we are guided by usage

scenarios provided by the medical experts. Therefore

we started from the actual data and modelled the

necessary concepts.

In our case, the diverse nature of medical data is

one of the main challenges for the formalization and

browsing of the patient records. We use ontologies

as a tool to link different scales and perspectives

(e.g. anatomical, cellular, behavioral etc.) and to

provide an abstract layer for structuring the

knowledge and the data to support efficient retrieval.

Another challenge for a successful multi-scale

diagnosis (where the data are acquired by different

acquisition sessions and modalities e.g. MicroCT,

MRI etc.), is to model these various kinds of data

representations according to different user’s

perspectives which could be quite diverse.

One of our main goals is to optimize the

visualization, search and browsing of the knowledge

base for the patient specific or general purpose data

that could assist in the clinical decision making

process. This will be supported by the development

of a computational framework that will take into

account the available morphological, structural, and

patient-specific information. There are two distinct

but complementary approaches to develop such a

framework: through standalone applications which

can support the demanding requirements of

computing resources and through a web-based

interface for sharing, browsing and searching the

data remotely.

4 PRELIMINARY RESULTS

In this section we present some work in progress

concerning the development of both standalone and

web-based applications for browsing and searching

the knowledge base, as well as annotating

segmented MRI scans and 3D models to facilitate

the discovery of heterogeneous medical data.

4.1 Standalone Applications

As mentioned in Section 3, our standalone

application provides a way to access the multi-modal

medical knowledge through the defined ontology.

The main objective of this application is to satisfy

various end-user perspectives (such as Radiologist,

Orthopedist, General practitioner, Tissue engineer)

for the early stage diagnosis of MSD.

In Figure 1, we provide a layered architectural

view of our proposed platform that supports the

HEALTHINF2014-InternationalConferenceonHealthInformatics

376

sharing and browsing of multi-modal, multi-scale

data. The knowledge management layer creates an

abstraction over the relational databases which

formalizes the structure of the medical data stored

and provides a way to access and update the data.

The system can support the retrieval of the multi-

scale data from the knowledge base by the use of

appropriate SPARQL queries, according to the usage

scenarios and competency questions formulated

from the requirement analysis phase. Some example

competency questions could be: Find all the data

related to Femoral cartilages of a patient

Mr.Brown who has a Musculoskeletal disease. I am

interested specifically in studying osteoarthritis.

Visualize the latest acquired 3D model of the

femoral cartilage of Mr.Brown. What is the

thickness of the cartilage? What was the thickness

of the cartilage one year ago? Show me the latest

Motion capture data (MoCap) of the patient. Show

me all the cells contained in the Femoral Cartilage.

Answers to the above competency questions can

be derived directly from the knowledge base and, in

addition, the platform can provide efficient ways to

explore this knowledge further to provide a mapping

between the data and the derived knowledge.

Figure 1: Proposed system architecture for accessing

multi-modal medical data using ontologies.

We have developed a user interface using the Jena

framework (Jena2 ontology API, online) which

provides a way to visualize and navigate through the

structured knowledge described by the ontology. It

can also support the browsing and searching of

patient data and visualization of the selected 3D data

(meshes and volumes).

The implemented prototype depicted on Figure 2

serves as a framework to ease the annotation

pipeline, by coupling manual or automatic

segmentation with automatic computation of

diagnostic measurements which may be applied to

the segmented 3D parts (such as volume, area etc.);

also, we foresee to develop methods for semi-

automatically or automatically add/modify the

semantic annotation according to the analysis of

inter-linked data of different scales, by exploiting

the power of multi-scale inference to aid medical

doctors in their data analysis processes and follow-

up studies. This is our first attempt to provide a way

to link the 3D geometry with knowledge by

attaching semantic information to 3D data.

Figure 2: Interface to access and annotate the patient data

along with formalized knowledge.

Another developed standalone application addresses

a different kind of knowledge regarding the

measurement of the femoral cartilage thickness.

Accurate and precise assessments of cartilage

thickness are important for addressing a number of

clinical questions for the prevention, treatment and

progression of osteoarthritis. For example, the

mechanical loading during walking has been shown

to influence the progression of osteoarthritis at the

knee as well as the outcome of treatment (Koo et.

al., 2005).

3D models of the femoral cartilage were created

from segmented magnetic resonance images, which

offer the potential of quantifying the cartilage

morphology with better accuracy than two-

dimensional plane images, and the weight bearing

regions of the cartilage that sustain contact during

walking were identified. The separation of weight

bearing regions from the non-weight bearing regions

of the knee joint is an important condition for the

study of osteoarthritis.

Using this tool, the cartilage thickness over each

region can be calculated and displayed as a color

map (see Figure 3). Focusing on the weight bearing

regions, which are usually of the greatest clinical

interest, this tool can assist in the progression

monitoring and affect the treatment outcome. The

outputs of this tool (3D models and measurements)

KnowledgeManagementinMedicine-AFrameworktoOrganize,BrowseandRetrieveMedicalData

377

are used in the knowledge management system.

4.2 Web-based Browsing, Sharing

and Searching using the Knowledge

Base

We intent to further exploit semantics and

knowledge management in a web based

environment, to support the integration of processes

within the medical investigation and the

visualization pipeline, utilizing scientific

visualization using WebGL for displaying 3D

models and medical data on the user’s web browser.

Figure 3: Tool for measuring the femoral cartilage

thickness near the weight bearing regions.

The conceptualization of medical content can be the

basis for more intelligent representations, with

intrinsic meaning (including contextual and domain

knowledge) and covering different user perspectives.

This way we will be able to identify, collect and link

all correlated information, making them accessible

to physicians during diagnostic and follow-up

processes.

Our goal is to create a web-based platform for

data exchange and knowledge sharing. The

ontology-driven knowledge management system

will support the organization and browsing of all

different kinds of stored medical content and

information, as well as different ways of searching

for these resources.

Three main search mechanisms will be provided:

(a) a keyword search for browsing and discovering

resources, (b) an advanced (semantic) search which

will utilize the knowledge base (e.g. SPARQL

queries combined with an OWL reasoner) and (c) a

geometric search mechanism, which will be based

on different kinds of similarity measures for shape

matching. Our aim is to offer an integrated way to

explore and visualize the data and support combined

search modalities to improve the retrieval

effectiveness.

Figure 4: Web-based data browsing and visualization.

We are planning to provide web interfaces for

information filtering/refinement and knowledge

visualization, as well as guided user interaction for

uploading, searching and navigation purposes. Some

preliminary work is shown in Figure 4.

5 DISCUSSION AND FUTURE

WORK

Nowadays CAD systems are already included in

medical imaging modalities such as digital

mammography, CT and MRI. Radiologists use this

type of CAD systems mainly for consulting purposes

before making their final decision thus reducing the

overall analysis time and manual efforts.

In order to assist in the diagnosis process, it

would be possible to search for and retrieve relevant

cases with a known pathology and compare the

therapy used in the past, which could increase the

physician's confidence in his/her decision. Of

course, this would require a storage system that

could host and logically organize a large number of

cases, and a reliable methodology and definition of

appropriate similarity measures. An intelligent

knowledge management platform could be able to

handle all of the above activities and interactions.

In addition, semantic search is essential for

connecting and exploiting this information.

Ontology management tools could support users in

maintaining and evolving knowledge models to meet

their needs. Finally, tools are needed to facilitate the

medical investigation process and help with the

annotation of images and 3D models.

Meaningful semantic annotations/associations

HEALTHINF2014-InternationalConferenceonHealthInformatics

378

can make the knowledge contained in medical

sources (such as MRI, X-rays, CT etc.) available in a

structured way, allowing both accurate and focused

retrieval and knowledge sharing. Moreover, this

knowledge can be used to provide valuable services;

for example, it could help the diagnostic procedure,

the therapy planning and all kinds of different

assessments by the medical team, tasks which

usually consume doctors’ valuable time.

Knowledge-based methods have an enormous

potential to manage, access and share the increasing

amount of visual information produced. Merging

ontologies (which provide a generic knowledge/

information framework) with computer-aided

diagnosis systems would result in a solution

targeting patient specific information. For example,

ontology-driven knowledge could be used to

improve the segmentation and analysis process as

well as the follow-up and treatment of a patient. In

addition, significant benefits can also be foreseen

regarding the visualization of the patient specific

data in a multi-scale, multi-modal and multi-

perspective environment.

In this context, we propose a platform composed

of loosely coupled components, either web-based or

standalone, that could support the medical

investigation process and could provide different

views on the data in a multi-scale collaborative

working environment. Our work intends to define a

framework for capturing invaluable expert

knowledge that is mostly undocumented or

implicitly contained in medical data, and therefore

hard to be reused or automated. Our goal is to foster

semantically augmented systems and services for

clinical decision-making, research and learning.

ACKNOWLEDGEMENTS

This work is supported by the FP7 Marie Curie

Initial Training Network "MultiScaleHuman: Multi-

scale Biological Modalities for Physiological Human

Articulation", contract MRTN-CT-2011-289897.

The authors would like to thank all the MSH

partners and especially Softeco Sismat S.r.l. and

CNR IMATI for their valuable help and support.

REFERENCES

AIM@SHAPE, FP6-IST Network of Excellence:

Advanced and innovative models and tools for the

development of semantic-based systems for handling,

acquiring, and processing knowledge embedded in

multidimensional digital objects. (Online), Available:

http://www.aimatshape.net (20 Oct. 2013)

Attene, M., Giannini, F., Pitikakis, M. and Spagnuolo, M.,

2013. The VISIONAIR Infrastructure capabilities to

support research. Computer-Aided Design and

Applications, Vol. 10, No.5, 2013, pp.851-862.

BioDigital Human (Online), Available:

https://www.biodigitalhuman.com (20 Oct. 2013)

Catalano, C. E., Robbiano, F., Parascandolo, P., Cesario,

L., Vosilla, L., Barbieri, F., Spagnuolo, M., Viano, G.,

and Cimmino, M. A., 2012. Exploiting 3D part-based

analysis, description and indexing to support medical

applications. In Workshop on Medical Content-based

Retrieval for Clinical Decision Support, in conjunction

with MICCAI, Nice, France.

Chen, L., Martone, M.E., Gupta, A., Fong, L. and Wong-

barnum, M., 2006. OntoQuest: Exploring Ontological

Data Made Easy. Very Large Data Bases - VLDB , pp.

1183-1186.

Doi, K., 2007. Computer-aided diagnosis in medical

imaging: historical review, current status and future

potential. Computerized Medical Imaging and

Graphics, 31: 198–211.

International Classification of Diseases, Ninth Revision,

Clinical Modification (ICD-9-CM). [Online],

Available: http://www.cdc.gov/nchs/icd/icd9cm.htm

(20 Oct. 2013)

Jena ontolgy API. (Online), Available:

http://jena.apache.org/index.html (20 Oct. 2013)

Koo, S., Gold, G.E., Andriacchi, T.P., 2005.

Considerations in measuring cartilage thickness using

MRI: factors influencing reproducibility and accuracy.

Osteoarthritis and Cartilage, 13: 782–789.

Magnenat-Thalmann, N., Gilles, B., Delingette, H.,

Giachetti, A. and Agus, M., 2007. Towards the Virtual

Physiological Human. In Eurographics 2007, Prague,

Czech Republic.

MyCF browser. (Online), Available:

http://www.mycorporisfabrica.org/ (20 Oct. 2013)

Palombi, O., Bousquet, G., Jospin, D., Hassan, S.,

Revéret, L. and Faure, F., 2009. My Corporis Fabrica:

a Unified Ontological, Geometrical and Mechanical

View of Human Anatomy. In 2nd workshop on 3D

Physiological Human, 3DPH2009, Zermatt,

Switzerland. LNCS, Spinger, Vol. 5903, pp. 207–219.

Pitikakis, M., Vavalis, M. and Houstis, C., 2012.

Semantically interoperable three-dimensional

scientific objects. The Knowledge Engineering Review

(KER), Vol. 27, No 01, 2012, pp. 33-55.

Rodríguez-González, A., Hernandez-Chan, G., Colomo-

Palacios, R, 2012. Towards an ontology to support

semantics enabled diagnostic decision support

systems. Current Bioinformatics. 7(3):234–245.

Rosse, C. and Mejino, JLV, 2003. A reference ontology

for biomedical informatics: the foundational model of

anatomy. J Biomed Inform, Vol. 36, pp. 478–500.

Spackman, K., 2000. Snomed rt and snomed ct. promise of

an international clinical terminology. MD Comput.,

Vol. 17(6):29.

Vasilakis, G., Garcia-Rojas, A., Papaleo, L., Catalano, C.

KnowledgeManagementinMedicine-AFrameworktoOrganize,BrowseandRetrieveMedicalData

379

E., Robbiano, F., Spagnuolo, M., Vavalis, M. and

Pitikakis, M., 2010. Knowledge-based representation

of 3D media. International Journal of Software

Engineering and Knowledge Engineering (IJSEKE),

Vol. 20, No. 05, pp. 739-760.

Virtual Soldier Project. (Online), Available:

http://www.virtualsoldier.net/summary.htm (20 Oct.

2013)

Voxel-Man. (Online), Available: http://www.voxel-

man.de/3d-navigator/inner_organs/ (20 Oct. 2013)

Zygote Body browser. (Online), Available:

http://www.zygotebody.com/ (20 Oct. 2013)

HEALTHINF2014-InternationalConferenceonHealthInformatics

380