NUCLEAR MEDICINE IMAGE MANAGEMENT SYSTEM

FOR STORAGE AND SHARING BY USING GRID SERVICES

AND SEMANTIC WEB

Daniela Giordano, Carmelo Pino, Concetto Spampinato

Department of Informatics and Telecommunication Engineering, University of Catania, 95125 Catania, Italy

Marco Fargetta

Italian Institute for Nuclear Physics (INFN), Catania, Italy

Angela Di Stefano

Institute of Neurological Science, Italian Research Council (CNR), Catania, Italy

Keywords:

Medical data sharing, Logical ﬁle catalogue (LFC), Metadata service (AMGA), SPECT, PET, Web 2.0.

Abstract:

Large amounts of images (SPECT, PET, scintigraphy) in the nuclear medicine ﬁeld have been routinely

produced in the last decades. In this paper we propose an image management system that allows nuclear

medicine physicians to share the acquired images and the associated metadata both locally (i.e. within the

same medical institute) and globally with other nuclear medicine physicians located anywhere in the world

by using GRID services for data (LFC) and metadata (AMGA) storage. The proposed system guarantees

medical data protection by anonymization that removes most sensitive data for unauthorized users, and

encryption, that guarantees data protection when it is stored at remote sites. Another important issue is that

often nuclear medicine data is associated with other medical data (e.g. neurological data) for diagnosis and

therapy follow-up. In order to correlate images with other clinical information, the common metadata are

enriched by developing a controlled vocabulary, which integrates known standards such as FOAF, CCR and

GeneOntology. All the metadata are stored in an RDF (Resource Description Framework) repository in order

to make the system fully compatible with existing metadata storage systems following the semantic web’s

philosophy.

1 INTRODUCTION

Usually, nuclear medicine physicians carry out ex-

aminations, e.g. Single Photon Emission Computed

Tomography (SPECT), Positron Emission Tomogra-

phy (PET) or scintigraphy, without a complete knowl-

edge of the clinical history of the patients. In fact,

most of the raw data collected in experiments or clin-

ical trials is usually stored either in different places

or in separate paper forms. Therefore there is a real

need of integrating heterogeneous data sources in or-

der to have a holistic vision of the patient’s health sta-

tus. Currently, existing methods, e.g. (Gabber et al.,

2003), (Xu and Jiang, 2009) barely address medical

data management needs beyond the speciﬁc depart-

ment’s boundaries, while it is known that the patient

medical folders are wide spread over many medical

sites involved in the patient’s healthcare. For this rea-

son, many efforts have been done in last years to de-

velop an interoperable infrastructure for digital health

((Cheung et al., 2009), (Freund, 2006)) by using se-

mantic web concepts, even if the attention is mainly

oriented to metadata sharing. A relevant project in

this direction is the e-Child project (Freund, 2006)

that develops a healthcare platform for pediatrics and

aims at integrating ontologies to homogenize biomed-

ical (from genomic, through cellular, disease, patient

and population-related) data. The system proposed in

this paper follows, and for some aspects, overcomes

the above medical data management’s approach be-

Giordano D., Pino C., Spampinato C., Fargetta M. and Di Stefano A..

NUCLEAR MEDICINE IMAGE MANAGEMENT SYSTEM FOR STORAGE AND SHARING BY USING GRID SERVICES AND SEMANTIC WEB.

DOI: 10.5220/0003170500800086

In Proceedings of the International Conference on Health Informatics (HEALTHINF-2011), pages 80-86

ISBN: 978-989-8425-34-8

 2011 SCITEPRESS (Science and Technology Publications, Lda.)

cause it permits to store images associating them with

not directly related metadata (i.e. not included in the

DICOM ﬁle) by developing an RDF controlled vocab-

ulary, which links the most common ontologies used

in the medical ﬁeld with personal health records on

the Internet (e.g. Google Health). This makes our

system interoperable with methods that share data fol-

lowing semantic web (e.g. in (Holford et al., 2009))

enriches the common information with personal data,

i.e. not present in medical facilities.

Given the huge amount of medical data involved in

this process and the need of collaboration among re-

searchers our system is provided with a GRID layer.

Many GRID systems have been proposed for han-

dling medical data geographically distributed on var-

ious medical centers, such as HOPE (Diarena et al.,

2008), MediGrid (Montagnat et al., 2005) or MAM-

MOGRID (Amendolia, 2004), but most of them deal

only with data ﬁles and do not provide higher level

services for manipulating medical data or for handling

the associated metadata, and also lack in the integra-

tion with other systems.

Therefore, in this paper we propose an approach for

sharing and analyzing nuclear medicine images where

1) metadata are stored in RDF, thus giving the possi-

bility to easily grab information coming from loca-

tions who share data according to the web semantic

approach and 2) image sharing among different med-

ical institutes is handled by GRID services in a trans-

parent way for the users. The key features of the pro-

posed system are:

• Full functionality in cases of network malfunc-

tioning is ensured since data are stored both lo-

cally (in the user’s computer and in the medical

institute server) and globally on GRID;

• It is interoperable with existing sharing methods

that use semantic web;

• It is ﬂexible since researchers can share content

choosing when and what to share;

• The metadata, usually contained in the DICOM

ﬁles, are enriched in order to take into account the

whole clinical history of patients. Users may de-

cide to export all the information (data and meta-

data), only data, only metadata or only part of it;

• Data encryption is implemented using SSH,

which ensures data integrity over Internet trans-

fers;

• Data privacy is guaranteed by removing sensitive

information before any transmission.

The paper is so organized: in section 2 the storage

and sharing system is described. In detail, we present

how the storage and the sharing work inside a medi-

cal institute using a client-server architecture and how

it works among different institutes using GRID ser-

vices. Sect. 3 and Sect. 4 describes, respectively, the

high level features for managing the interoperability

with other systems, for image analysis and for system

querying and the user interface. Finally, concluding

remarks are given.

2 STORAGE AND SHARING

SYSTEM OVERVIEW

In order to develop a distributed environment for im-

age and information sharing to support the diagnosis,

the treatment of patients and for statistical evaluation,

the system is provided with two levels of storage and

sharing: the ﬁrst is locally managed by a client-server

architecture, deployed in the medical institute where

the nuclear medicine physicians belong to, whereas

the second one is on GRID and allows global data

sharing, i.e. data may be shared among different in-

stitutes using the services offered by the GRID com-

puting. Fig. 1 shows the architecture of the proposed

system for the local and global data sharing.

The typical use case is the following: a nuclear

medicine physician stores the images and the meta-

data of a performed examination in its own local

database (located in his/her computer). Afterwards,

the client creates an anonymous version of the data re-

moving all the conﬁdential information so they can be

sent to the main server in the respect of privacy issues.

Additionally, the client allows users to deﬁne the set

of metadata he/she wants to share both in GRID and

in his/her medical institute.

The data transmission between client and server runs

asynchronously in order 1) to make the system robust,

in fact in case that, Internet connection is unavailable,

data are locally stored and subsequently sent to the

main server and to GRID when the connection will

be available again and 2) to keep users unaffected by

the actual time needed for the data transfer.

The server is provided with repositories where all

the metadata produced within the same institute are

stored. The communication with GRID is delegated

to the server, thus optimizing the bandwidth’s use.

2.1 Local Data Storage and Sharing

Inside a medical institute, data are stored and shared

using a standard client-server architecture, as shown

in ﬁg. 2. The client and the server are connected by a

local network or a VPN (Virtual Private Network).

The client contains the user interfaces and implements

the logical communication with the GRID infrastruc-

ture. It also contains a ﬁle repository (for image stor-

NUCLEAR MEDICINE IMAGE MANAGEMENT SYSTEM FOR STORAGE AND SHARING BY USING GRID

SERVICES AND SEMANTIC WEB

Figure 1: Architecture for Local and Global Data Storage and Sharing.

age) and a SESAME server (Broekstra et al., 2001)

(for RDF metadata storage), in order to save patient’s

data locally.

The server also includes a ﬁle repository and a

SESAME metadata repository for the data produced

by all the nuclear medicine physicians within the

same institute. Data are sent from the client to the

server using File Transfer Protocol (FTP), whereas

metadata is transmitted to SESAME server using

Simple Object Access Protocol (SOAP) requests by

means of a webservice, as shown in ﬁg. 2.

Figure 2: Local Data Storage and Sharing.

By using the client interface, a nuclear medicine

physician can record and manage patients, add infor-

mation to patient’s clinical history (according to the

schema shown in the next section), include any rele-

vant documents (textual reports, generic images, DI-

COM images, etc..), run queries locally or on GRID

data, associate the metadata deriving from the queries

to the data locally stored and perform statistical anal-

ysis on sets of data and virtual data (i.e. coming from

the main institute center or from GRID).

2.2 Global Data Sharing on GRID using

LFC and AMGA

Data sharing among different institutes, geograph-

ically distributed, is implemented on GRID and it

is based on the paradigm to create virtual environ-

ments where large amount of data and complex com-

putations can be performed by different communi-

ties grouped in Virtual Organisation (VO). Usually,

each VO offers several services for GRID participants

in order to simplify the data management providing

them basic functionality to store and retrieve ﬁles.

The access to the GRID is hidden by a middleware.

In this work we used the EGEE Grid and the G-Lite

middleware

. The two main services of the G-lite

middleware for data and metadata storage, used in

our system, are: the Logical File Catalogue (LFC)

(Venugopal et al., 2004) and the AMGA Metadata

Service (Nuno, 2006). The LFC allows users to as-

sociate a logical name to a ﬁle in a hierarchy for-

mat like a local ﬁle system, hiding the real location

of the storage. Moreover, a logical name may refer

many replicas, so a user can retrieve the ﬁle through

its logical name from the nearest location in a trans-

parent way. The AMGA Metadata Service is a special

database designed to store metadata associated with

ﬁles. Therefore, its internal structure reproduces a ﬁle

system hierarchy where the directories are collections

of metadata deﬁned in a custom schema and each ﬁle

in the directory is an entry containing the values for

the metadata. This approach allows to easily map a

ﬁle name with a set of metadata inside AMGA.

EGEE website - http://www.eu-egee.org/ and gLite

Grid middleware website -http://www.glite.org/

HEALTHINF 2011 - International Conference on Health Informatics

The communication with GRID infrastructure is man-

aged by the medical institute server, which also aims

at maintaining aligned the information stored in nu-

clear medicine physicians’ computers and the one

stored in GRID. In each institute the main server

represents the bridge between clients and GRID, as

shown in ﬁg. 3.

Figure 3: Global Data Storage and Sharing.

The ﬂow diagram of the interaction between a

generic medical institute and the GRID infrastructure

is shown in ﬁg. 4, when a nuclear medicine physician

requires the storage of a speciﬁc image on GRID by

using a proper GUI (Graphical User Interface). The

steps performed by our system are: 1) the client sends

to the server the proxy certiﬁcate (needed for the ac-

cess to GRID) previously created and the identiﬁer of

the image to be stored on GRID, 2) the server of the

medical institute (for simplicity called proxy server)

queries the ﬁle repository to check if the ﬁle is present

or not, if not it asks the client to send the image, then

it sets the necessary permissions to read, write for the

GRID, 3) the server queries the GRID database using

the medical institute Grid ID given by the client in or-

der to see if the image was already stored, if not 4) the

server removes sensitive data, sends it to a Grid Stor-

age Element, writes in the LFC catalog an appropri-

ate logical name and writes (uploads if the image was

previously stored) the metadata in the AMGA server.

3 HIGH LEVEL FEATURES

In order to provide useful information about the stored

images and to make them available with the related

Figure 4: Interaction with Medical Institute - GRID Infras-

tructure.

metadata to the nuclear medicine community, the sys-

tem is provided with high level features. More in de-

tail, the system contains three processing levels:

• A semantic layer that enriches patient metadata

by a controlled vocabulary using RDF/XML stan-

dard. This level guarantees the interoperability

with existing frameworks;

• An image processing layer that analyzes the

stored images. This is an important layer, since

sometimes is desirable to share only the process-

ing results and not the entire image. This level

performs the image analysis and interacts with the

semantic layer for the storage of the processing re-

sults in RDF/XML;

• Query Composition for performing complex

queries both locally and on GRID. This module

allows users to search useful information by pro-

cessing only the metadata available locally or in

GRID.

The interaction between the three levels and the

system’s architecture is shown in ﬁg. 5.

3.1 Semantic Layer

Usually nuclear medicine images (e.g. in PET,

SPECT) are stored in DICOM format, containing the

metadata provided with the standard. These metadata

are not sufﬁcient for describing the clinical history

of patients. For this reason we enrich the informa-

tion available in order to give the nuclear medicine

physicians the possibility to better ﬁgure out a spe-

ciﬁc disease by developing a model that represents

the medical data so that it can be analyzed by seman-

tic tools. In detail, the system stores concepts, speci-

ﬁes typed relationships between these concepts using

NUCLEAR MEDICINE IMAGE MANAGEMENT SYSTEM FOR STORAGE AND SHARING BY USING GRID

SERVICES AND SEMANTIC WEB

Figure 5: Interaction between the high level features’ sys-

tems.

RDF

(Resource Description Framework) with XML

syntax format. More in detail, we enrich the DICOM

metadata by developing a controlled vocabulary that

includes:

• Personal data by using FOAF ontology

;

• Generic Health Information according to the Con-

tinuity of Care Record (CCR) standard (Detmer

et al., 2008) such as: diagnoses, allergies, medi-

cation list, immunizations, family history, social

history, vital signs, procedures, symptoms, plan

of care, functional status, biosignals (EEG, ECG,

etc...);

• Genetic information using GeneOntology (Ash-

burner, 2000);

• Neurological detailed information by using Mesh

(Soualmia et al., 2004);

• Image processing information that represents the

output of the implemented image processing algo-

rithms and which introduces a new semantic level

to the stored metadata.

An example of RDF ﬁle for a patient is shown

in ﬁg. 6. It is notable that this information is in-

serted by the users, but it can be easily obtained by

http://www.w3.org/RDF/

http://www.foaf-project.org/

Figure 6: Example of produced RDF File.

querying systems that share data using RDF. For in-

stance, personal data in FOAF can be derived from a

generic social network or by using a v-card, whereas

generic health information can be obtained by the

user’s Google Health Account

or other systems that

aim at storing online health care data such as the one

proposed in (Bielikov

a and Moravc

ık, 2008).

Metadata storage has been carried out by using

SESAME server (Broekstra et al., 2001) so that these

information may be available also for other purposes.

The sensitive data, such as Name, Surname, SSN

must be available only for the nuclear medicine physi-

cian who carries out the examination, and are not ex-

ported in RDF in order to ensure data privacy.

3.2 Image Processing Layer

This level is provided with a set of processing meth-

ods for medical image analysis. The output of this

processing is stored according to the semantic layer

and is related to the speciﬁc processed image. This

allows users to also share the results of the processing

avoiding to send the original images when it is not re-

quired, resulting in less bandwidth usage.

The implemented methods for nuclear image analy-

sis, also available for MRI, X-Rays images, are:

• Measurement of distances, angles and some pa-

rameters within the images;

• The contrast absorption curve over time;

https://www.google.com/health/

HEALTHINF 2011 - International Conference on Health Informatics

• Image Texture and Image Contour Analysis for

speciﬁc organs;

• Pattern recognition for identifying brain struc-

tures.

Example of such algorithms have been proposed

by the authors in (Faro et al., 2010a), (Faro et al.,

2010b) and (Giordano et al., 2009). Therefore, when

a user performs one of the above methods, the out-

put is treated as metadata and stored in the SESAME

server.

3.3 Query Composition

The query composition level aims at building com-

plex queries both locally and on GRID. The queries

are performed only on the metadata (stored in the

SESAME server and in the AMGA server) since a

content based image retrieval module is not present.

This level receives the user’s queries (by using a con-

trolled GUI) and interacts both with the local storage,

performing SPARQL queries on the SESAME server,

and with the GRID, where queries are performed fol-

lowing the approach proposed in (Montagnat et al.,

2008).

4 CLIENT INTERFACE

The client interface consists of a multiform Java ap-

plication, allowing users to manage any type of infor-

mation concerning the patients. The functionalities

available for nuclear medicine physicians are:

• view information about their patients ;

• search a patient by typing the name;

• insert, modify or delete a patient;

• print a patient’s report;

• show the patient exams;

• insert, modify or delete an exam;

• management of medical history;

• view and process DICOM ﬁles;

• export the patient’s data on GRID;

• query the system both locally and globally on

GRID.

For each functionality a speciﬁc GUI has been de-

signed. For example, ﬁg. 7 shows the main interface

where a list of patients and a tool to manage patients’

data is available. In the bottom area, there is also a

section to handle the exams for each patient. Fig. 8,

instead, shows the interface for DICOM ﬁles and im-

age processing.

Figure 7: Main GUI.

Figure 8: GUI for DICOM File Processing.

Moreover, for expert users the client application

contains a form (ﬁg. 9) to set parameters for the FTP,

GRID and AMGA.

Figure 9: GUI for parameters’ setting.

NUCLEAR MEDICINE IMAGE MANAGEMENT SYSTEM FOR STORAGE AND SHARING BY USING GRID

SERVICES AND SEMANTIC WEB

5 CONCLUDING REMARKS

In this paper we presented a distributed system of

data and metadata storage, sharing and processing.

Web semantic data modeling using RDF permits that

all the information can be stored and reused. The

RDF modeling allows us to efﬁciently integrate het-

erogeneous data into a coherent whole and to provide

description of data elements. The area of applica-

tion, is still conﬁned to the speciﬁc sector of nuclear

medicine but the ﬂexibility of the architecture permits

the application to be used on a greater number of med-

ical ﬁelds by a simple re-modeling. The platform will

be integrated with an image retrieval system in order

to assist the physicians in the retrieval of images by

content. A future development is to provide the sys-

tem with an ontology layer that will map high level

requests to low level queries.

REFERENCES

Amendolia, S. R. (2004). Mammogrid: A service oriented

architecture based medical grid application. In GCC,

pages 939–942.

Ashburner, M. (2000). Gene ontology: Tool for the uniﬁca-

tion of biology. Nature Genetics, 25:25–29.

Bielikov

a, M. and Moravc

ık, M. (2008). Modeling the

reusable content of adaptive web-based applications

using an ontology. In Advances in Semantic Media

Adaptation and Personalization, pages 307–327.

Broekstra, J., Kampman, A., and Harmelen, F. V. (2001).

Sesame: An architecture for storing and querying rdf

data and schema information. In Semantics for the

WWW. MIT Press.

Cheung, K. H., Prud’hommeaux, E., Wang, Y., and

Stephens, S. (2009). Semantic Web for Health Care

and Life Sciences: a review of the state of the art.

Brief. Bioinformatics, 10:111–113.

Detmer, D., Bloomrosen, M., Raymond, B., and Tang, P.

(2008). Integrated personal health records: Transfor-

mative tools for consumer-centric care. BMC Medical

Informatics and Decision Making, 8(1):45.

Diarena, M., Nowak, S., Boire, J. Y., Bloch, V., Donnar-

ieix, D., Fessy, A., Grenier, B., Irrthum, B., Legre,

Y., Maigne, L., Salzemann, J., Thiam, C., Spalinger,

N., Verhaeghe, N., de Vlieger, P., and Breton, V.

(2008). HOPE, an open platform for medical data

management on the grid. Stud Health Technol Inform,

138:34–48.

Faro, A., Giordano, D., Pino, C., Spampinato, C., and

Di Stefano, A. (2010a). 3D striatum reconstruction

of 123ioﬂupane MRI images for quantitative assess-

ments on the dopaminergic neurotransmission system.

In IEEE International Workshop on Medical Mea-

surements and Applications (MeMeA 2010), Ottawa,

Canada.

Faro, A., Giordano, D., Spampinato, C., and Pennisi, M.

(2010b). Statistical texture analysis of MRI images to

classify patients affected by multiple sclerosis. In 12th

Mediterranean Conference on Medical and Biological

Engineering and Computing MEDICON 2010, pages

236–239, Porto Carras, Chalkidiki, Greece. Interna-

tional Proceedings of the IFBME, Springer.

Freund, J. (2006). Health-e-child: an integrated biomedical

platform for grid-based paediatric applications. Stud

Health Technol Inform, 120:259–270.

Gabber, E., Fellin, J., Flaster, M., Gu, F., Hillyer, B.,

Ng, W. T.,

Ozden, B., and Shriver, E. A. M. (2003).

Starﬁsh: highly-available block storage. In USENIX

Annual Technical Conference, FREENIX Track, pages

151–163.

Giordano, D., Scarciofalo, G., Spampinato, C., and

Leonardi, R. (2009). Automatic skeletal bone age as-

sessment by integrating EMROI and CROI process-

ing. In IEEE International Workshop on Medical

Measurements and Applications (MeMeA 2009), Ce-

traro, Italy.

Holford, M. E., Rajeevan, H., Zhao, H., Kidd, K. K., and

Cheung, K. H. (2009). Semantic Web-based inte-

gration of cancer pathways and allele frequency data.

Cancer Inform, 8:19–30.

Montagnat, J., Breton, V., and Magnin, I. (2005). Partition-

ing medical image databases for content-based queries

on a Grid. Methods Inf Med, 44:154–160.

Montagnat, J., Frohner,

A., Jouvenot, D., Pera, C., Kun-

szt, P., Koblitz, B., Santos, N., Loomis, C., Texier, R.,

Lingrand, D., Guio, P., Rocha, R. B. D., de Almeida,

A. S., and Farkas, Z. (2008). A secure grid medical

data manager interfaced to the glite middleware. J.

Grid Comput., 6(1):45–59.

Nuno, N. S. (2006). Distributed metadata with the amga

metadata catalog. In Workshop on NextGeneration

Distributed Data Management - HPDC-15.

Soualmia, L., Golbreich, C., and Darmoni, S. (2004). Rep-

resenting the mesh in owl: Towards a semi-automatic

migration. In Proceedings of the KR 2004 Work-

shop on Formal Biomedical Knowledge Representa-

tion, pages 81–87.

Venugopal, S., Buyya, R., and Winton, L. (2004). A grid

service broker for scheduling distributed data-oriented

applications on global grids. In MGC ’04: Proceed-

ings of the 2nd workshop on Middleware for grid com-

puting, pages 75–80, New York, NY, USA. ACM.

Xu, Z. and Jiang, H. (2009). Hass: Highly available, scal-

able and secure distributed data storage systems. Com-

putational Science and Engineering, IEEE Interna-

tional Conference on, 2:772–780.

HEALTHINF 2011 - International Conference on Health Informatics