This article was published in an Elsevier journal. The attached copyis furnished to the author for non-commercial research and

education use, including for instruction at the author’s institution,sharing with colleagues and providing to institution administration.

Other uses, including reproduction and distribution, or selling orlicensing copies, or posting to personal, institutional or third party

websites are prohibited.

In most cases authors are permitted to post their version of thearticle (e.g. in Word or Tex form) to their personal website orinstitutional repository. Authors requiring further information

regarding Elsevier’s archiving and manuscript policies areencouraged to visit:

http://www.elsevier.com/copyright

Author's personal copy

INFORMATION TECHNOLOGY REVIEW

MammoGrid d a prototype distributedmammographic database for Europe

R. Warrena,*, A.E. Solomonidesb, C. del Fratec, I. Warsia, J. Dinga,M. Odehb, R. McClatcheyb, C. Tromansd, M. Bradyd, R. Highname,M. Cordelle, F. Estrellab, M. Bazzocchic, S.R. Amendoliaf

aDepartment of Radiology, University of Cambridge, Addenbrooke’s Hospital, Cambridge, bCCCSResearch Centre, University of West of England, Frenchay, Bristol, UK, cInsitituto di Radiologia,Universita di Udine, Italy, dDepartment of Engineering Science, Oxford University, Oxford,eSiemens Molecular Imaging, Oxford, UK, and fETT Division, CERN, 1211 Geneva 23, Switzerland

Received 12 May 2006; received in revised form 18 September 2006; accepted 29 September 2006

This paper describes the prototype for a Europe-wide distributed database of mammograms entitled MammoGrid,which was developed as part of an EU-funded project. The MammoGrid database appears to the user to be a singledatabase, but the mammograms that comprise it are in fact retained and curated in the centres that generated them.Linked to each image is a potentially large and expandable set of patient information, known as metadata. Transmis-sion of mammograms and metadata is secure, and a data acquisition system has been developed to upload and down-load mammograms from the distributed database, and then annotate them, rewriting the annotations to thedatabase. The user can be anywhere in the world, but access rights can be applied. The paper aims to raise awarenessamong radiologists of the potential of emerging ‘‘grid’’ technology (‘‘the second-generation Internet’’).ª 2007 The Royal College of Radiologists. Published by Elsevier Ltd. All rights reserved.

Introduction

The increasing use of electronic formats for radio-logical images, including mammography (the par-ticular focus of the authors), together with thefast, secure transmission of images and patientdata, potentially enables many hospitals andimaging centres throughout Europe to be linkedtogether to form a single ‘‘virtual organization’’.The technological possibilities are co-evolving withan appreciation of potential uses. Huge ‘‘feder-ated’’ databases of mammograms, which appearto the user to be a single database, but are in factretained and curated in the centres that generatedthem, has been tested as a useful model. Linked to

each image would be a potentially large andexpandable set of relevant information, known asmetadata. This might comprise: patient age, ex-ogenous hormone exposure, family and clinicalhistory, even genetic information; informationknown to correspond to risk factors (e.g., diet,parity, breast density); as well as image acquisitionparameters, including breast compression andexposure data, which affect image appearance.Levels of access to the images and metadata in thedatabase would vary according to the certificatedrights of the user: radiologists might have access toall of it, whereas epidemiologists and researcherswould have more limited access, protecting pa-tient privacy, in accordance with European legis-lation. Such databases might consist of millions ofmammograms providing a common resource ofimages and metadata that might be used fora variety of purposes, such as education, research,and access to expert second reading.

* Guarantor and correspondent: R. Warren, University of Cam-bridge, Department of Radiology, Cambridge Breast Unit, Box97, Addenbrooke’s Hospital, Cambridge CB2 2QQ, UK. Tel.:þ44 1223 586959; fax: þ44 1223 217886.

E-mail address: rmlw2@cam.ac.uk (R. Warren).

0009-9260/$ - see front matter ª 2007 The Royal College of Radiologists. Published by Elsevier Ltd. All rights reserved.doi:10.1016/j.crad.2006.09.032

Clinical Radiology (2007) 62, 1044e1051

Author's personal copy

Currently, such virtual organizations and hugefederated databases do not exist. However, theenabling technology to realize them does. It isvariously known as the ‘‘second-generation inter-net’’, or the ‘‘grid’’.y The grid may be understoodwith reference to the Internet. The Internet pro-vides access to a massive number of files, each ofwhich has a unique resource locator (URL), orweb address used to locate that file. There arealmost no restrictions on what the content maybe (or the accuracy) of files posted on the Inter-net. Also, although methods for secure financialtransactions are now available, the Internet is no-toriously insecure d most computer viruses aredistributed via it. Despite these limitations, useof the Internet has grown explosively over recentyears. At its technical heart, the Internet com-prises a remarkably small set of software standardsand protocols that have been universally adopted.z

The grid is, similarly, an emerging set of softwarestandards that builds upon those developed forthe Internet, to provide for consultation of filesand for distributed computing. The name derivesfrom an analogy with the electrical power grid.The grid aims to provide computing power inmuch the same way. Like the Internet, it is notnecessary to know how the grid works in order touse it. It is sufficient for this paper to accept thefollowing assertions: (1) the grid is an emergingset of standards, not yet completely agreed; and(2) it has the potential to provide the technologicalinfrastructure to realize the virtual organizationand federated database concepts.

The EU funded project ‘‘MammoGrid’’ set out toexplore the following conjecture: grid technologyand standards have evolved to the point wherea prototype, federated database of mammogramsmight be constructed, based on centres in threeEuropean countries (UK, Italy, Switzerland; Fig. 1).More specifically, as part of the MammoGrid pro-ject, the following parameters were explored:

Image standardization

As noted above, the appearance of a mammogramis greatly affected by differences in image-acquisition processes (machine type, filter, exposure

time, etc). As part of the MammoGrid project, thepossibility of standardizing images using theStandard Mammogram Form (SMF) representation,developed by Highnam and Brady,1 to support theapplications of research study was explored. TheGenerate-SMF program used in this project is in-stalled on the ‘‘grid box’’ employed between thepartner organizations of the MammoGrid pro-ject.2,3 The SMF program supports two areas ofresearch by the clinical partners in the collabora-tion: (1) Breast density as a risk factor. Mammo-graphic density is recognized as a major riskfactor for breast cancer.4 The SMF program providesa measure of the amount and proportion of densetissue, so that partners in Cambridge and Udinecould compare breast density measurements pro-vided by SMF with the standard method of visualassessment5 and an automated, two-dimensional(2D), interactive computer program.6 (2) Com-puter-aided detection of microcalcifications andmasses. Before MammoGrid, the partners fromSassari and Pisa had developed a system named‘‘CALMA’’ (computer-aided library for mammogra-phy) for the detection of lesions and microcalcifi-cations with good sensitivity and specificity, aspreviously reported.7,8 The database of mammo-grams generated during the MammoGrid projectat Cambridge and Udine was used to re-assessthe sensitivity and specificity of CALMA, and toexamine whether its performance would be im-proved using the standardized images generatedby SMF.

The present study tests the use of the databasefor research in three areas: (1) epidemiologicalstudies of breast cancer risk using mammographicdensity estimated by a computer method; (2) totrain pattern recognition algorithms, such as thoseused for computer-aided detection (CAD). Thevariability of breast anatomy and imaging condi-tions (machine type, exposure time, tube voltage,filter, etc) imply the need for a huge number oftraining cases for pattern-recognition algorithms.9

For data mining applications (‘‘FindOneLikeIt’’)in which a radiologist queries the database tofind images (or a region of interest in an image)that resemble the current case and for which thediagnosis is biopsy proven.

Material and methods

The project has approval from the Cambridge andNorfolk research ethical committees (LREC; twoseparate submissions). UK women participatingin one subproject where heights and weightswere obtained gave written informed consent to

y The word ‘‘grid’’ as used in this paper should not be con-fused with the anti-scatter grid familiar to all mammographers.The two uses of the same word is unfortunate, but there is norelation at all between the two meanings.z The software standards are: html - hypertext markup lan-

guage; http - hypertext transfer protocol; and tcp/ip: transmis-sion control protocol & internet protocol. The familiar www isa particular network application that uses tcp/ip, and http en-ables resources to be communicated over the Internet.

A prototype distributed mammographic database for Europe 1045

Author's personal copy

participate. Italian women gave consent by agree-ing to informed participation, which involved sub-mitting to height and weight measurement andallowing their data to be used for research. TheMammoGrid project has approval from the Cam-bridge and Norwich LRECs (2002) for storage ofanonymized images with associated metadata inan encrypted format for viewing between theproject partners, and for cancers and controlsfrom screening they did not require informedconsent to be obtained in view of the anonymousnature of the project and the soundness of theencryption process.

The MammoGrid project has several technicalaspects: (1) image standardization using SMF; (2)the development of a workstation on which imagescan be acquired, annotated, and uploaded to thegrid; (3) the distribution of data, images, andclinician queries across grid-based databases. Forthe UK, ethical approval was obtained for use ofanonymized data, but patient consent was notrequested. The Italian participants did not requireany ethical application. Ethical, legal, confidenti-ality, and security constraints apply differently inthe partner countries of origin, and must berespected.

The aim has been to use, wherever possible,freely available software to make queries acrossa widely distributed, federated database of mam-mographic images, and to perform epidemiologicalstudies and CAD on the sets of returned images.For example, in the MammoGrid project, radiolo-gists may annotate (i.e., mark out) differentregions of a mammogram, which are then sub-jected to different CAD algorithms (includingCALMA) and compared with stored mammogramsin the database. As any one of these stages maybe executed independently or take some timeto be completed, the process must be controlledin a way that recognizes the current state ofthe computation and ensures that results aremeaningfully assembled from the various partialoutcomes. To provide for these possibilities,MammoGrid has built on the results of previousEuropean grid projects, such as EGEE.10e12 The de-tails of EGEE and how it was used in MammoGridhave been described elsewhere.13

To be clinically useful, it is evident that theMammoGrid system has had to attain high levels ofdata integrity, quality, and consistency: theserequirements have been met (by developing stan-dard services, standard data formats, and strict

Figure 1 Schematic of the MammoGrid project. Four sites are shown contributing mammograms to the distributeddatabase, at Udine, CERN (Geneva), and Cambridge. An additional node is shown at Siemens Molecular Imaging (for-merly CTI Mirada) in Oxford, where the MammoGrid data acquisition system was developed. One node, not shown onthe diagram, is located in Oxford University.

1046 R. Warren et al.

Author's personal copy

automated quality checks.) Once approval hasbeen obtained from the relevant ethical commit-tee on behalf of patients whose anonymized datawill be held in the database, the system designenables certification and authorization services toguarantee anonymity and security. These criticallyimportant technical issues are describedelsewhere.13

Image standardization: SMF

The SMF technology is a fully automated, objectivemeasurement tool to estimate the volume ofglandular tissue in the breast from a mammo-gram.1,14 The SMF algorithm explicitly considersbreast compression, exposure and tube voltage,and computes two volumetric measures of breastdensity, (1) the absolute volume (cm3) of thebreast that is dense (SMF volume) and (2) the per-centage of the volume of the breast that is dense(SMF%). SMF is different from other volumetric re-search methods in that it incorporates a full phys-ics model rather than using step-wedges in eachimage.9,15 It has been described in two recentstudies.16,17 It is also used in the MammoGrid pro-ject to standardize the images on the grid box be-fore using the CALMA CAD program to test whetherimproved detection follows.

MammoGrid data acquisition workstation

A screenshot of the MammoGrid data acquisitionworkstation is shown in Fig. 2. The MammoGridworkflow currently begins with films, which mustbe digitized. The digitized film, with correspondingmetadata, then needs to be uploaded, on the fed-erated database so that it can subsequently bedownloaded for processing. Subsequent processingmay involve a radiologist supplying annotations,and there may be several such annotations, eitherdone serially, or independently. Alternatively, pro-cessing may involve computer algorithms, eitherGenerateSMF to compute density, or to detect mi-crocalcifications and/or masses. Again, such pro-cesses may be applied serially or independently.The results of all such processing need to be auto-matically uploaded to the database, with appropri-ate guarantees of security, confidentiality, andwith an appropriate audit trail. Achieving all ofthese aims has resulted in the MammoGrid data ac-quisition workstation.

In the case that a radiologist examines a mam-mogram, in order to supply or revise an annotation,the case needs to be displayed. The developmentof a high-quality retrieval and mammogram display

system was not a major concern of the MammoGridproject, as such display systems are availablealready from a range of manufacturers.18 A rudi-mentary system was developed to the extent thatthe distributed database concept could be evalu-ated and the clinical studies accomplished.

User requirements

As in any major software project, the MammoGridpartners were confronted by a dilemma: becauseall the end-uses of the database were unknown thesystem was not fully developed, thus enablingalterations as new challenges and requirementswere presented.

The methodology based on ‘‘use-cases’’, [skel-etal scenarios used to describe standard interac-tions between actors (roles enacted by users) anda given system] were used to determine the limitsof that system (the system scope). This approachwas adopted in MammoGrid to model and docu-ment the workflow of a radiology department.19

This publication concludes that information infra-structures to support radiology not only must ad-dress issues related to the integration of clinicaldata from heterogeneous databases, but mustfacilitate access and filtering of patient data. Anexample of a use-case from the MammoGrid pro-ject, namely ‘‘perform radiological analysis’’ isshown in Fig. 3 [reproduced from 20].

DICOM

The MammoGrid project conforms to the DICOMstandard21 in two respects. First, the digitized im-ages are imported and stored in the DICOM storageformat, so that the full set of image and patient-related metadata is readily available with the im-ages, and that information exchange with othermedical devices understanding the DICOM storageformat is seamless. To ensure the compatibilitywith DICOM conformant clients, it is also necessaryto exchange DICOM datasets via a communicationprotocol also defined by the standard.

CRISTAL and metadata

CRISTAL11 is a distributed scientific database systemused in high-energy physics experiments at CERN, TheEuropeanLaboratory forParticlePhysics.TheCRISTALproject has studied the use of a description-drivenapproach using metadata-modelling techniques tomanage the evolving data needs of a large commu-nity of scientists. This approach has been shown toprovide many powerful features such as scalability,

A prototype distributed mammographic database for Europe 1047

Author's personal copy

system evolution, interoperability, and reusability,aspects that are essential for future-proofing medi-cal information systems. The MammoGrid projectwas based largely on the conjecture that CRISTALwould be ideally suited to being grid-enabled, inorder to serve as the basis of a MammoGrid queryhandler.

The handling of a grid database query is illus-trated in Fig. 4.12 A query-handling tool locally cap-tures the elements of a clinician’s query and issuesa formal query, using appropriate grid software,against the metadata structures and users’ struc-tured terms, across multiple data centres in thedistributed hospitals. At each location, the queriesare resolved against relevant metadata and con-stituent sub-queries are remotely executed onthe local mammogram databases. Selected sets ofmatching mammograms may then be either ana-lysed remotely or replicated back to the centre at

which the clinician issued the query for subsequentlocal analysis, depending on the philosophy adop-ted in the underlying grid software, but maintain-ing the impression that a single database hasbeen accessed, as if resident at the local machine.

Joining centres together: ‘‘grid boxes’’

At the launch of the MammoGrid project, a soft-ware system named ‘‘AliEn’’ had been developedat CERN to provide a ‘‘virtual file catalogue’’ thatenables transparent access to distributed data-sets. The AliEn software was installed and config-ured on a set of novel secure hardware units,nicknamed grid boxes. The idea has been that eachMammoGrid centre would have a single point ofentry onto the MammoGrid (Fig. 1). Each grid boxis responsible for storing new patient imagesand studies, updating the file catalogue, and

Figure 2 The MammoGrid data acquisition system, developed at Siemens Molecular Imaging (formerly CTI Mirada),in Oxford. A typical screenshot is shown (a) and an enlarged version of the user-interaction buttons shown in (b). Theaim of this part of the project was not to develop a state-of-the-art visualization workstation, rather to develop onethat could interact with the grid, to enable uploading/downloading of mammograms and annotations.

1048 R. Warren et al.

Author's personal copy

propagating any changes in such a way that theview of database at the workstations is always upto date at every site.

Results

At the final EC review, the ability of the system toachieve complex queries across the grid to thesatisfaction of technical reviewers, appointed bythe EC, was demonstrated and confirmed asa functioning system. The experimental findingshave been submitted separately to journals rele-vant to the clinical hypotheses; the findings of themammography data have been published in thisissue of Clinical Radiology.22

Discussion

MammoGrid is one of a number of EuropeanHealthGrid projects.23,24 The technical approachesvary, but e-Diamond in the UK and MammoGrid inEurope have adopted the grid as their platform

of choice proof of principle of the concept. Asbreast screening in the UK and in Italy has beenbased on film, mammograms have had to be digi-tized for use in both e-Diamond and MammoGrid.The UK ‘‘e-Science’’ programme has funded thee-Diamond project,25,26 but whereas MammoGridis based on open-source software, e-Diamond isbased on (IBM) proprietary technology and concen-trates on two complementary applications, namelyteaching26 and ‘‘FindOneLikeIt’’. Also, in theUnited States, the National Digital MammographyArchive27 (NDMA) has adopted a radically differentapproach, using a large centralized archive of di-rect digital mammograms.

The central feature of the MammoGrid projectis a geographically distributed, grid-based data-base of standardized images and associatedpatient data. The novelty of the MammoGridapproach lies in the application of grid technologyand in the provision of data and tools, whichenable radiologists to compare new mammogramswith existing ones on the grid database, allowingthem to make comparative diagnoses.

Figure 3 Typical use-case hierarchy and associated diagram. See text for details.

A prototype distributed mammographic database for Europe 1049

Author's personal copy

This project demonstrates the potential of thedatabase, populated with provenance-controlled,reliable data from across Europe, and provides theprospect of statistically robust epidemiology, al-lowing analysis of lifestyle factors, including, forexample, diet, exercise, and exogenous hormoneuse. A grid-based system such as MammoGridwould also be suitable for storing genetic orpathological image information. The project hasattracted attention in the computer community asa paradigmatic exemplar of the application of gridtechnology. Although the problems required toinstall such systems have not yet been overcome,the project has established an approach and a pro-totype platform, sharing medical data, includingimages, across a grid. In loose collaboration witha number of other European medical grid projects,it is addressing the issues of informed consent andethical approval, data protection, compliance withinstitutional, national and European regulations,and security.12,28

References

1. Highnam R, Brady M. Mammographic image analysis. 1st ed.Dordrecht: Kluwer Academic Publishers; 1999.

2. Estrella F, McClatchey R, Rogulin D, et al. A service- basedapproach for managing mammography data. Stud HealthTechnol Inform 2004;107:235e9.

3. Amendolia SR, Brady M, McClatchey R, et al. MammoGrid:large-scale distributed mammogram analysis. Stud HealthTechnol Inform 2003;95:194e9.

4. Boyd NF, Rommens JM, Vogt K, et al. Mammographic breastdensity as an intermediate phenotype for breast cancer.Lancet Oncol 2005;6:798e808.

5. Warner E, Lockwood G, Tritchler D, et al. The risk of breastcancer associated with mammographic parenchymal pat-terns: a meta-analysis of the published literature to

examine the effect of method of classification. Cancer De-tect Prev 1992;16:67e72.

6. Byng JW, Yaffe MJ, Lockwood GA, et al. Automated analysisof mammographic densities and breast carcinoma risk.Cancer 1997;80:66e74.

7. Bazzocchi M, Facecchia I, Zuiani C, et al. Application ofa computer-aided detection (CAD) system to digitalizedmammograms for identifying microcalcifications. RadiolMed (Torino) 2001;101:334e40.

8. Bottigli U, Delogu P, Fantacci ME, et al. Search of Microcal-cification clusters with the CALMA CAD station. Int Soc OptEng (SPIE) 2002;4684:1301e10.

9. Pawluczyk O, Augustine B, Yaffe M, et al. A volumetricmethod for estimation of breast density on digitizedscreen-film mammograms. Med Phys 2003;30:352e64.

10. Robert Jones, et al. The Information Societies TechnologyProject. In: EU-EGEE, EU Contract IST-2003e508833,http://www.eu-egee.org/. Page accessed 22nd [May 2007].

11. Meta-Data Objects as the Basis for System Evolution F.Estrella, Z. Kovacs, J-M. Le Goff & R. McClatchey. LectureNotes in Computer Science; vol. 2118, pp. 390e399 ISBN:3-540-42298-6 Springer-Verlag, 2001 (Presented at the 2ndACM SIGMOD Int Conf on Web-Age Information Management(WAIMaV�2001). XiaV�an, China. July 2001.

12. Estrella F, Hauer T, McClatchey R, et al. A grid informationinfrastructure for medical image analysis. In: Gibaud BDM,editor. Proceedings of the International Workshop onDistributed Databases and Processing in Medical ImageComputing; 2004; Rennes, France 2004.

13. Amendolia SR, Estrella F, Del Frate C, et al. Deployment ofa grid-based medical imaging application. Stud HealthTechnol Inform 2005;112:59e69.

14. Marias K, Behrenbruch C, Highnam R, et al. Amammographic image analysis method to detect and mea-sure changes in breast density. Eur J Radiol 2004;52:276e82.

15. Berks M, Diffey J, Hufton A, et al. ‘‘Feasabiliaty and acc-eptability of stepwedge based measurement’’. In:Astley S, editor. International Workshop on DigitalMammography. Berlin: Springer Verlag; 2006. p. 355e61.

16. Jeffreys M, Warren R, Highnam R, et al. Initial experiencesof using an automated volumetric measure of breast den-sity: the standard mammogram form. Br J Radiol 2006;79:378e82.

Figure 4 Handling a query in the MammoGrid distributed database system. See text for details.

1050 R. Warren et al.

Author's personal copy

17. Highnam R, Pan X, Warren R, et al. Breast composition mea-surements using retrospective standard mammogram form(SMF). Phys Med Biol 2006;51:2693e713.

18. Evertsz CJG, Bodicker A, Bohnenkamp S, et al. Soft-copyreading environment for screening mammography-Screen.In: Yaffe MJ, editor. Digital Mammography. The 5th Inter-national Workshop on Digital Mammography, IWDM 2000,Toronto, Canada.MadisonWI,USA: MedicalPhysicsPublishing;2000. p. 566e72.

19. Bui AA, Taira RK, Dionisio JD, et al. Evidence-based radiol-ogy: requirements for electronic access. Acad Radiol 2002;9:662e9.

20. Odeh M, Hauer T, McClatchey R, et al. Use-case driven ap-proach in requirements engineering: the MammoGrid Pro-ject. In: Hamza MH, editor. Proceedings of the 7th IASTEDInt. Conference on Software Engineering & Applications;2003; Marina del Rey, CA. ACTA Press; 2003. p. 562e7.ACTA Press Calgary, Canada.

21. The DICOM standard. In: Digital imaging and communica-tions in medicine. p. http://medical.nema.org/. ACCESSEDMay 2005, authors National Electrical ManufacturesAssociation.

22. Warren R, Thompson D, del Frate C, et al. A comparisonof some anthropometric parameters between anItalian and a UK population: ‘‘proof of principle’’ of a

European project using MammoGrid. Clin Radiol 2007;62:1052e60.

23. Amendolia SR, Estrella F, del Frate C, et al. Deployment ofGrid-based medical imaging application. Stud Health Tech-nol Inform 2005;112:59e69.

24. Nørager S, Paindaveine YT. The HealthGrid terms of refer-ence. Brussels: EU; 2002.

25. Brady M, Gavaghan M, Simpson A, et al. eDiamond: a grid-enabled federated database of annotated mammograms.In: Berman F, editor. Grid computing: making the global in-frastructure a reality. Chichester, UK: Wiley; 2003. p.923e43.

26. Brady M, Gilbert F, Lloyd S, et al. eDiamond: the UK’s digitalmammography national database. In: Pisano E, editor.International Workshop on Digital Mammography. NC,USA: Chapel Hill; 2004.

27. The National Digital Mammography Archive. http://www.ndma.us/pdf/NDMA_brochure_2006.pdf (accessed on 23rdMay 2007).

28. From Grid to HealthGrid Editors: T Solomonides, RMcClatchey, V Breton, Y Legre & S Norager. Studies inHealth Technology & Informatics vol. 112, ISBN: 1-58603-510-X, ISSN: 0926-9630 IOS Press. (Proceedings the 3rdHealthGrid International Conference (HGaV�05). Oxford,UK. April 2005).

A prototype distributed mammographic database for Europe 1051