Managing Provenance for Medical Datasets

An Example Case for Documenting the Workﬂow for Image Processing

Ajinkya Prabhune

, Rainer Stotzka

, Michael Gertz

, Lei Zheng

and J

urgen Hesser

3,4

Institute for Data Processing and Electronics, Karlsruhe Institute of Technology, Eggenstein-Leopoldshafen, Germany

Database Systems Research Group, Institute of Computer Science, Heidelberg University, Heidelberg, Germany

Experimental Radiation Oncology, Medical Faculty Mannheim, Heidelberg University, Mannheim, Germany

Experimental Radiation Oncology, IWR, Heidelberg University, Heidelberg, Germany

Keywords:

Scientiﬁc Data Repository, ProvONE Provenance Standard, Open Annotation Data Model, DICOM Dataset,

Angioscopy Workﬂow, Metadata Management.

Abstract:

In this paper, we present a novel data repository architecture that is capable of handling the complex image

processing workﬂows and its associated provenance for clinical image data. This novel system has unique and

outstanding properties versus existing systems. Among the most relevant features are a ﬂexible and intuitively

usable data and metadata management that includes the use of a graph-based provenance management strategy

based on a standard provenance model. Annotation is supported to allow for ﬂexible text descriptors as being

widespread found for clinical data when structured templates are not yet available. The architecture presented

here is based on a modern database and management concepts and allows to overcome the limitations of

current systems namely limited provenance support, lacking ﬂexibility, and extensibility to novel requests. To

demonstrate the practical applicability of our architecture, we consider a use case of automated image data

processing workﬂow for identifying vascular lesions in the lower extremities, and describe the provenance

graph generated for this workﬂow. Although presented for image data, the proposed concept applies to more

general context of arbitrary clinical data and could serve as an additional service to existing clinical IT systems.

1 INTRODUCTION

Electronic health records (EHR) offer a digital doc-

umentation of the diagnostic and therapeutic history

of a patient. Parts of these records are managed by

Hospital information systems (HIS) and subsystems

like radiology information systems (RIS). Over the

initiative integrating the healthcare enterprise (IHE)

deﬁned workﬂows enable standardized records on one

hand side and (as a ﬁnal goal) a complete coverage of

all procedures in a clinic. However, from the perspec-

tive of data provenance, these systems partially solve

the data management problem. Data provenance is

hereby of utmost relevance since it allows traceabil-

ity with respect to validation and reproducibility of

diagnostic and therapeutic procedures and decisions

(Estrella et al., 2007). However, despite its relevance,

this topic is scarcely discussed for clinical routine

data reporting with some exceptions in bioinformat-

ics (Davidson et al., 2007). Approaches that have

been widespread used are neither complete, nor ﬂexi-

ble and are therefore difﬁcult to integrate in a clinical

environment.

In the following, we propose a new architecture

for medical data repository system with a dedicated

focus on medical image data processing. Currently,

the image data is mostly obtained in radiology depart-

ments, and the acquired data such as Magnetic Reso-

nance Imaging (MRI) or Computerized Tomography

(CT) data is typically stored in a Picture Acquisition

and Communication System (PACS), backed by DI-

COM 3.0 format (Mildenberger et al., 2002). This

format allows storing image data and the associated

metadata such as patient id, type of acquisition, date,

etc, that is embedded within the DICOM ﬁles. Thus,

all relevant information required to repeat the acqui-

sition is stored and accessible via reading and inter-

preting the DICOM header ﬁles.

Although being in widespread use, two major lim-

itations of the PACS are, (a) the lack of a modern

database (NoSQL) technology for storing, indexing

and accessing metadata. (b) the brittleness of han-

236

Prabhune A., Stotzka R., Gertz M., Zheng L. and Hesser J.

Managing Provenance for Medical Datasets - An Example Case for Documenting the Workﬂow for Image Processing.

DOI: 10.5220/0006109402360243

In Proceedings of the 10th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2017), pages 236-243

ISBN: 978-989-758-213-4

dling the workﬂow and provenance information for

reproducing the results. For example, manipulations

of image data via image processing routines (which

will dominate in the future to automatically process

standard cases), the processed result is stored as new

DICOM data, even if only small modiﬁcations on the

header information were performed.

From the perspective of repeatability, the image

processing workﬂow would have to be stored in the

DICOM data as well. This can be accomplished either

by using private tags or by deﬁning a new modality.

The former step is an ad-hoc solution since there are

no formal rules how this should be done and gener-

ally we cannot assume that all users are handling this

issue with sufﬁcient care. Hence, there is no guaran-

tee that this image processing workﬂow can ever been

repeated. The latter strategy requires a new DICOM

modality standard, which is an immense overhead.

From this perspective, there is a demand for an

easy-to-use technology that not only maps all meta-

data formats, workﬂows, and patient related data but

also enables users to describe new workﬂows and still

guarantee the provenance information.

This lack of ﬂexibility of the current system and

the brittleness of the DICOM standard led us to pro-

pose a different type of data repository architecture

that is overcoming these limitations. Instead of intro-

ducing a disruptive solution in the existing PACS in-

frastructure, we present an auxiliary system that pro-

vides the following functionality;

• Metadata management for extracting, modelling,

indexing and storing the metadata embedded from

DICOM-ﬁle in a ﬂexible and a scalable database

• Provenance tracking using standard provenance

models such as ProvONE and PREMIS

for the

image processing workﬂows.

• Data annotation (image annotations) for systematic

capturing of vital details in the standard Open An-

notation Data Model (Sanderson et al., 2013).

• Data preservation for allowing long-term access

and reusability of the data.

• Data quality control and data curation for speciﬁc

tools controlling rules of good practice and later

diagnostic and treatment guidelines.

2 IMAGE PROCESSING DATA

REPOSITORY

The goal of the Image Processing Data Repository

(IPDR) is to provide the various auxiliary functional-

http://www.loc.gov/standards/premis/

!"#$%&#'#"()*+,'-*.

/%%*&0 &#*%(!$+1#"$2

30&0(3#2"*1$+4 567-#"0&#*%(!$+1#"$

8$&090 &0(:;&+0"&#*%

8$&090 &0(5+*"$22#%<

8$&090&0(8*9$--#%< 8$&090&0(!&*+0<$

30&0(5+$2$+10&#*%

30&0(/%0-42#2

30&0(=6+0&#*% 30&0(5+*"$22#%<

>53?(

!$+1#"$2

8$&09 0&0(

80%0<$@$%&

30&0(

80%0<$@$%&

/""$22(A04$ +(B /5>

30&0(C+0%2'$+

>%<$2&D3*.%-*09(

)*+,'-*.2

/6&E$% &#"0&#*%(0%9(

/6&E*+#F0&#*%

G02$(8$&090&0(

?$<#2&+0&#*%

>%&$+0"& #1$()$7(5*+&0-

H%*.-$9<$(?$I+$2$%&0&#*%

J2$+B

>%&$+'0"$

K5=(=-62&$ +

L/II-#"0&#*%(K*2&#%<(

!42&$@M

8$&090&0(!&*+0<$

30&0(/+"E#1$(

L5/=!(2$+1$+M

>@0<$(5+*"$22#%<(30&0(?$I*2#&*+4(

N$%$+#"(=-#$%&

Figure 1: Image Processing Data Repository Architecture.

ity for handling the lifecycle of DICOM dataset stored

in the PACS server. The IPDR follows a modular ar-

chitecture design, as shown in Figure 1. The modular

design of IPDR allows us to extend the functionality

to fulﬁl any new requirements put forth by the radia-

tion therapy (RT) research community. The complete

IPDR architecture is developed in Java, the various

REST services are implemented in Jersey (Sandoval,

2009) and the front-end is developed in Vaadin frame-

work. The IPDR is deployed on a web server (Tomcat

7). Following is the description of the core compo-

nents of the IPDR architecture.

2.1 Generic Client

The Generic Client component provides a convenient

solution for the RT medicine and research commu-

nity to integrate their existing software/tools with the

IPDR seamlessly. The Generic Client is an upgrade

to the GCS-API (Prabhune et al., 2015) that was pre-

viously developed for handling the datasets of the

nanoscopy research community. The Generic Client

is extended to handle the DICOM ﬁles. The various

modules of Generic Client are explained below:

• Access Layer-API: The access layer API exposes

the various interfaces for connecting to the existing

radiation therapy data processing tools. Currently,

we have developed a command line interface (CLI)

over the access layer API that allows the RT com-

munity to transfer the datasets.

• Base Metadata Registration: For long term

archival, it is necessary to register of the DICOM

ﬁles in IPDR. However, the DICOM ﬁles main-

Managing Provenance for Medical Datasets - An Example Case for Documenting the Workﬂow for Image Processing

237

tain its proprietary metadata schema that needs to

be translated to the standard metadata schema sup-

ported by the IPDR. This module enables the au-

tomated registration of the DICOM datasets by ex-

tracting the administrative metadata concerning the

DICOM Study, Series, and Image and mapping it

to the Core Scientiﬁc Metadata Model (CSMD)

(Matthews, 2015).

• Ingest/Download Workﬂows: Ingest/Download

workﬂows module holds the predeﬁned workﬂow

that allow the ingestion of DICOM data from client

systems to the IPDR or to download the DICOM

datasets from the IPDR to the client system.

• Data Transfer: The data transfer module allows

the transfer of DICOM datasets bi-directional from

multiple endpoint, i.e. from client system or data

acquisition systems to IPDR or vise versa. Cur-

rently, the HTTP WebDAV (Goland et al., 1999)

protocol is supported for high-throughput transfer

of the datasets. The various interfaces for inte-

grating other transfer protocols such as, FTP and

gridFTP (Allcock et al., 2005) are available. The

data transfer module is designed to be fail-safe. For

example, data transfers that are interrupted are au-

tomatically re-triggered for transfer. To optimize

high-volume data transfers, the number of parallel

WebDAV connections can be manipulated.

• Authentication and Authorization: The Generic

Client follows a two-fold process. First, to enable

the registration of DICOM datasets , for a regis-

tered member of the RT research community, an

OAuth-secured RESTful connection is established.

Each researcher is provided with a unique access

token and access token secret. Second, for transfer-

ring the data, the WebDAV protocol authentication

is required. Currently, WebDAV authentication for

an entire research community is conﬁgured.

2.2 Image Processing Data Repository

The IPDR is a multi-module architecture that is sep-

arated into functionality speciﬁc module. The IPDR

is deployed as a web application with access to the

high-performance computing cluster, the large-scale

data storage, and a dedicated metadata database. Fol-

lowing is the description of each module of the IPDR:

IPDR Services: The services module provides the

various arbitrary high-end services for interacting

with datasets stored in IPDR.

• Scientiﬁc Workﬂow (medical Image Data Process-

ing workﬂow) submodule offers the integration

of a workﬂow engine for automating the execu-

tion of the image processing workﬂows. Cur-

rently, the Apache ODE

BPEL workﬂow engine

is integrated with IPDR and workﬂows deﬁned

in BPEL speciﬁcation (OASIS, 2007) are sup-

ported. The Prov2ONE (Prabhune et al., 2016) al-

gorithm handles the automatic creation of prospec-

tive ProvONE graph, which is stored in a graph

data model of the metadata storage.

• Data discovery submodule provides the users to

search the ingested DICOM datasets based on the

metadata stored in the metadata storage. The meta-

data are indexed for enabling full-text search, for

searching over provenance graphs various queries

implemented as RESTful services are provided.

Finally, to enable large scale metadata harvesting

the Open Archives Initiative Protocol for Metadata

Harvesting (OAI-PMH)

is implemented.

• Annotation service is an interactive submodule that

allows enriching of the images through the anno-

tation service. Valuable information provided in

form of annotations by the researchers is modeled

using the Open Annotation Data Model and main-

tained in the metadata storage.

• Publication service submodule allows users to

share (provide open access) the experiment dataset

with other research communities, which can be

based on data exchange technologies like i2b2

Metadata: The metadata module is responsible for

handling all the metadata speciﬁc tasks in the IPDR.

Metadata can either be embedded inside the DICOM

ﬁles or for other ﬁle formats it can be ingested sep-

arately with the datasets. For enabling querying,

sharing and reusing of metadata, DICOM metadata

needs to be stored in a dedicated metadata storage.

Furthermore, as DICOM metadata is subject to fre-

quent changes throughout its lifecyle, it is necessary

to have a ﬂexible database model. Thus, we chose

ArangoDB

, a database which offers three different

types of data models, namely, key-value, document

and graph data model.

• Metadata Extraction: The various metadata ex-

tractors for extracting the metadata from DICOM,

HDF5, TIFF and XML ﬁles are provided as inde-

pendent micro-services by this module. For ex-

tending towards other ﬁle formats, this module

provides a generic interface that can be imple-

mented by any new metadata extractors.

• Metadata Modelling: The various metadata mod-

els (schemas) are registered through this com-

http://ode.apache.org/

https://www.openarchives.org/pmh/

www.i2b2.org

https://www.arangodb.com/documentation/

HEALTHINF 2017 - 10th International Conference on Health Informatics

238

ponent. For example, the ProvONE provenance

model, CSMD, Metadata Encoding and Transmis-

sion Standard (METS)

and PREMIS are some of

the currently supported metadata schemas. Fur-

thermore, application-speciﬁc metadata model of

DICOM is also registered by this submodule.

• Metadata Processing: The metadata processing

submodule provides the handling and assembling

of the community speciﬁed METS proﬁle. METS

is metadata container format that comprises of var-

ious sections which allow encoding of adminis-

trative hamdSeci, structural hﬁleSeci, hstructMapi,

hstructLinki descriptive hdmdSeci and provenance

hdigiprovMDi metadata. The heterogeneous meta-

data comprising descriptive and administrative

metadata from the ArangoDB document store and

provenance metadata from graph store are assem-

bled in the METS proﬁle for enabling sharing of

entire metadata for a dataset.

• Metadata Storage: The various database speciﬁc

storage adapter for storing and querying the meta-

data are implemented in this submodule. The

CRUD operations for document metadata repre-

senting the contextual information from the DI-

COM dataset, for graph metadata representing the

provenance and workﬂows, and the six verbs of

the OAI-PMH protocol are implemented in Arango

Query Language (AQL).

Data: The Data module provides the integration with

the low-level functionality that is responsible for stor-

ing the data in the cache storage and further in the tape

storage for long term archival. This data storage rep-

resents the PACS server, which is enriched with the

high-performance computing (HPC) cluster for en-

abling processing of the DICOM dataset, where these

data processing services (algorithms) are deployed on

the HPC. Following are functionalities that extend the

underlying PACS server.

• Data Preservation: The Data Preservation submod-

ule provides with the checksum of the dataset that

is to be ingested in IPDR. This checksum is main-

tained in the PREMIS metadata schema for verify-

ing the consistency of a ﬁle during data transfers

through Generic Client.

• Data Analysis and Curation: The various com-

munity speciﬁc data analysis and curation algo-

rithm are registered in these submodules. The data

processing algorithms are deployed as individual,

reusable processes that expose a unique REST end-

point, which is used when assembling a scien-

tiﬁc workﬂow (necessary for composing the BPEL

workﬂow).

http://www.loc.gov/standards/mets/

• Data Processing: The integration with the HPC

cluster and conﬁguration of the execution envi-

ronment is handled by this submodule. More-

over, when new intermediary datasets are gener-

ated as a result of an execution of data processing

algorithm, the automated registration and ingest of

these datasets is performed by this submodule.

2.3 User Interface

A web-based user interface developed in Vaadin is

integrated with the IPDR. Feature such as free-text

search over metadata and faceted search are avail-

able for discovering the datasets. Provenance graphs

stored in ArangoDB are visually represented using the

D3.js (Zhu, 2013) framework. The OAI-PMH meta-

data services are exposed through the user interface

for harvesting the metadata. As the user interface

is integrated with the workﬂow engine, the radiation

therapy research community can perform remote exe-

cution of their image processing workﬂows.

3 DATA AND METADATA

WORKFLOW

Datasets in interventional radiological clinics follow

a systematic workﬂow, starting from the image acqui-

sition in DICOM format to the ﬁnal generation of a

treatment record. During each step of the workﬂow,

the DICOM images are subject to image manipula-

tions and diagnostic information is extracted; these

datasets are hence enriched with essential metadata

describing the diagnostic details at each step. More-

over, additional and related datasets might be created

as well. To manage these datasets, we described the

complete ﬂow of the DICOM datasets and metadata

beginning from the Data Acquisition system (DAQ)

and the treatment to making it accessible to clinicians

for ﬁnding similar cases or for quality assurance rea-

sons; and to use it for documentation. The complete

workﬂow, described in our example case consists of

eight steps as shown in Figure 2. (1) The workﬂows

begins when the either raw DICOM dataset acquired

from the CT scanner system is made available to the

Generic Client. (2) The Generic Client is entirely au-

tomated, it performs the registration of the dataset to

be ingested by extracting the base metadata from the

DICOM metadata section, translates it into the CSMD

standard and registers the dataset. (3) A successful

registration of base metadata triggers the transfer of

data to the PACS server. (4) The complete metadata

from the DICOM dataset is extracted, modelled and

Managing Provenance for Medical Datasets - An Example Case for Documenting the Workﬂow for Image Processing

239

Figure 2: Flow of data and metadata from clinical data acquisition and treatment system to IPDR.

validated. (5) The metadata is ingested to the ded-

icated metadata storage database. (6) Metadata is

segregated as descriptive or provenance metadata and

stored either in document data model or graph data

model of ArangoDB. (7) The DICOM dataset from

the cache storage is transferred to archive storage for

long-term preservation. The preservation metadata

associated with the dataset is updated in the meta-

data storage. (8) The metadata is indexed for enabling

free-text search and allowing discovery of the datasets

from the IPDR. The data and metadata is accessible in

the Vaadin based user interface which is connected to

the IPDR through various REST services.

4 WORKFLOW AND

PROVENANCE

A successful completion of the Schedule (Acquisi-

tion) Workﬂow (SWF) generates the various DICOM

ﬁles that are stored in the PACS server. The Post-

Processing Workﬂow (PAWF) is the logical extension

of the SWF (Liu and Wang, 2010), which aims for de-

riving additional qualitative and quantitive data that is

beneﬁcial for improving the patient’s treatment. Typ-

ically, in a clinical environment, there are two cate-

gories of post-processing workﬂows.

Distributed application (agent) oriented work-

ﬂows: In the case of distributed agent oriented work-

ﬂows, various information systems are controlled un-

der the authority of different health care actors such

as physicians, general practitioners and various hos-

pital departments. Currently, distributed agent ori-

ented post-processing workﬂow steps involves appli-

cations, such as Computer Aided Detection (CAD),

Image processing, 3D reconstruction and surface ren-

dering are available. However, these workﬂows are

often distributed, and the various applications partici-

pating in the execution of the workﬂow are deployed

on a stand-alone hosting system. Various endeavors

address the handling of provenance and workﬂows

in these distributed agent oriented workﬂows (Kifor

et al., 2006) (Zhang et al., 2009).

Custom data processing workﬂows: In the case of

custom data processing workﬂows the various data

processing steps (algorithms) are typically deﬁned

and implemented for advanced data processing in the

clinical environment. Currently, to coordinate and

trace the execution of the workﬂow steps, the work-

list embedded in the DICOM ﬁle is referred. The

DICOM worklist (to-do lists) hold the list of tasks

that are to be performed on DICOM datasets. How-

ever, DICOM does not have the capability to model

and maintain the provenance traces associated with

the data. Even though the worklist offers a conve-

nient technique for maintaining the workﬂow steps,

it obviously lacks an accepted workﬂow and prove-

nance standard. For modelling comprehensive prove-

nance information, i.e. both prospective provenance

(workﬂow execution plan) and retrospective prove-

nance (runtime events) (Zhao et al., 2006), the W3C

ProvONE standard is adopted. Furthermore, to en-

able reuse of these data processing algorithms, they

are registered as web-services in the Data Analysis

and Curation sub-module of the IPDR. The interme-

diate data generated after each step in the workﬂow

is ingested in the IPDR PACS server, thus preventing

unnecessary repetition of the workﬂow.

Using an existing angioscopy image analysis

workﬂow (Maksimov et al., 2009) (Brockmann et al.,

2010), the various processing steps are deployed as

HEALTHINF 2017 - 10th International Conference on Health Informatics

240

Figure 3: Angioscopy workﬂow provenance in ProvONE.

independent web services that are accessible through

a REST URI. Based on these REST URIs, the com-

plete workﬂow is modelled in BPEL speciﬁcation and

executed on the high-performance computing cluster.

Before the execution of the workﬂow, the ProvONE

prospective graph is generated by the Prov2ONE al-

gorithm, and during the execution enriched with the

runtime provenance information (retrospective prove-

nance). The complete ProvONE provenance graph for

the angioscopy workﬂow is shown in Figure 3.

During the execution of the image analysis work-

ﬂow the intermediate result are ingested into the IPDR

and the metadata is extracted, modelled and stored in

the metadata storage, as shown in the Figure 2.

5 DISCUSSION

In the available systems, clinical process data is stored

in diverse clinical information systems, for example

in HIS including subsystems such as a RIS and PACS.

The essence of these systems is not only the reporting

but also to be able to retrospectively follow the course

of the clinical decision and treatment. Whenever im-

age data is manipulated such as via image processing

or conclusions are driven via machine learning tech-

niques, there is no adequate tool available yet that al-

lows tracking these steps in an adequate way. The

rationale for considering the IPDR is its high ﬂexibil-

ity for different clinical processes and their adequate

Managing Provenance for Medical Datasets - An Example Case for Documenting the Workﬂow for Image Processing

241

documentation.

Furthermore, the beneﬁt of being able to track pre-

vious diagnostic and treatment processes is to be able

to assess the treatment outcome or the correctness of

the diagnostic step. This allows both to increase the

internal quality of clinical routines but also offers the

chance to get a deeper insight when e.g. for cancer

patients there is a relapse, which might offer different

ways of further patient treatment.

However, to handle the frequent metadata changes

in the DICOM dataset that are introduced during the

clinical studies, we integrated a database system sup-

porting multiple ﬂexible data models. Additionally,

the metadata is automatically indexed for enabling

free-text search through the data discovery service.

For capturing the dynamic information in the form

of annotations, the W3C standard Open Annotation

Data Model is implemented through the annotation

service. The IPDR allows the entire modelling of

the DICOM metadata in METS format and the ret-

rospective provenance is modelled in PREMIS stan-

dard, thus allowing efﬁcient sharing (harvesting) of

the metadata using the OAI-PMH services. For han-

dling the provenance of the image processing work-

ﬂows, we integrated the ProvONE standard in IPDR,

wherein both the workﬂow plan as well as the asso-

ciated provenance are captured, thus, enabling repro-

ducibility of scientiﬁc results.

We also presented the data and metadata workﬂow

describing the integration of IPDR with the PACS

server, see Figure 2. Our aim was to seamlessly in-

tegrate the entire IPDR into the existing PACS server

and network infrastructure without any disruption in

the existing execution environment. The DICOM

dataset either from the data acquisition system (imag-

ing modalities) or from the various workstations in the

PACS network is processed through IPDR and for-

warded to the PACS server. To handle the DICOM

data the open-source DICOM DCM4CHE (Zeilinger

et al., 2010) toolkit is integrated into the IPDR.

Regarding the regulations, documents concerning

diagnostic and therapeutic procedures have to be ﬁled

for about 10-30 years, documents for quality control

have to be archived for about 5 years depending on

country and regulations. There is no formal require-

ment for a detailed documentation of medical pro-

cesses so far but the availability of a ﬂexible prove-

nance software technique would foster regulation pro-

cedures if publicly available.

6 RELATED WORK

Currently, there are various commercial as well as

custom implementations of PACS solutions available

for clinical studies for handling the medical datasets.

Enterprise Imaging Repository (EIR) (Bian et al.,

2009) is an alternative to the commercial PACS server

and network solution that provides handling of DI-

COM ﬁles using the DCM4CHE toolkit. Various

functionality like HL7(Dolin et al., 2006) interfaces,

web access to DICOM objects and media creation ser-

vices (CDW) are implemented by the EIR.

For handling the cardiology datasets (Marcheschi

et al., 2009) have a PACS solution completely built

using open-source technology. This solution is based

on existing commodity hardware with a server farm

comprising a storage pool of 24 TB for saving the

cardiology datasets. A PACS based solution with au-

tomated workﬂow management techniques using the

YAWL speciﬁcation in radiology information system

(Zhang et al., 2009) are implemented for allowing

a workﬂow-aware and ﬂexible integration of various

components in building RIS.

The integration of a workﬂow system does of-

fer novel approach for a ﬂexible RIS design. How-

ever, the above mentioned PACS solutions fail to

provide many of the critical aspects associated with

the handling of the complete lifecycle of the med-

ical datasets. The IPDR solution presented in this

paper overcomes these limitations by providing func-

tionalities such as, metadata handling based on stan-

dard metadata models, dedicated scalable and ﬂexible

metadata storage, annotation services based on stan-

dard Open Annotation Data Model, integration with

high-performance computing cluster for performant

execution of post-processing workﬂow tasks and cap-

turing of provenance in ProvONE model.

7 CONCLUSION

In this paper, we presented the detailed architecture

and the functionalities provided by the IPDR, which

is a comprehensive repository system for handling the

EHR datasets. The IPDR is based on the principle

of modular architecture that enables easy extensibility

by adding task-speciﬁc modules for handling any new

requirements. The functionality provided by IPDR

enables the RT researchers to: (a) automatically ex-

tract, store, and access metadata in standard metadata

models, (b) allow reproducibility of scientiﬁc results

by capturing provenance in ProvONE standard, (c)

enrich the data quality by capturing annotations in the

Open Annotation Data Model, (d) enable the repeata-

HEALTHINF 2017 - 10th International Conference on Health Informatics

242

bility of complex image processing workﬂows using

integrated workﬂow engine, (e) large-scale metadata

harvesting through standard OAI-PMH protocol.

To demonstrate the handling of image processing

workﬂow with capturing of its associated provenance,

the angioscopy workﬂow modelled in BPEL was ex-

ecuted using a workﬂow engine integrated with the

IPDR, and the associated provenance was automati-

cally modelled in ProvONE (Figure 3).

However, the data exchange approach uses safe

data transfer strategies but are not yet adapted to clin-

ical needs. An integration of i2b2 and a data ware-

house that collects all clinical data in combination

with technologies such as PCORnet

offers the poten-

tial to standardised data exchange for studies and con-

sultation and will thereby allow using our architecture

to be integrated into a clinic-wide information system

of the next generation. In particular, this strategy is

considered to be integrated into data integration cen-

tres that are currently planned at the university med-

ical centre Mannheim (UMM) and the MIRACUM

consortium that is currently funded by the German

Ministry of Research and Education (BMBF). The

technology is built upon developments of the BMBF-

project LSDMA

REFERENCES

Allcock, W., Bresnahan, J., Kettimuthu, R., et al. (2005).

The globus striped gridftp framework and server. In

Proceedings of the 2005 ACM/IEEE conference on

Supercomputing, page 54. IEEE Computer Society.

Bian, J., Topaloglu, U., and Lane, C. (2009). Eir: En-

terprise imaging repository, an alternative imaging

archiving and communication system. In 2009 Annual

International Conference of the IEEE Engineering in

Medicine and Biology Society, pages 2168–2171.

Brockmann, C., Jochum, S., Hesser, J., et al. (2010). Graph-

matching-based computed tomography angiography

in peripheral arterial occlusive disease. Clinical Imag-

ing, 34(5):367 – 374.

Davidson, S. B., Boulakia, S. C., Eyal, A., Lud

ascher, B.,

McPhillips, T. M., Bowers, S., Anand, M. K., and

Freire, J. (2007). Provenance in scientiﬁc workﬂow

systems. IEEE Data Eng. Bull., 30(4):44–50.

Dolin, R. H., Alschuler, L., Boyer, S., Beebe, C., Behlen,

F. M., Biron, P. V., and Shabo (Shvo), A. (2006).

Hl7 clinical document architecture, release 2. Jour-

nal of the American Medical Informatics Association,

13(1):30–39.

Estrella, F., Hauer, T., McClatchey, R., Odeh, M., Rogulin,

D., and Solomonides, T. (2007). Experiences of en-

www.pcornet.org

www.miracum.de

www.helmholtz-lsdma.de

gineering grid-based medical software. International

Journal of Medical Informatics, 76(8):621 – 632.

Goland, Y., Whitehead, E., Faizi, A., Carter, S., and Jensen,

D. (1999). Http extensions for distributed authoring–

webdav. Technical report.

Kifor, T., Varga, L. Z., Vazquez-Salceda, J., Alvarez, S.,

Willmott, S., Miles, S., and Moreau, L. (2006). Prove-

nance in agent-mediated healthcare systems. IEEE In-

telligent Systems, 21(6):38–46.

Liu, Y. and Wang, J. (2010). PACS and digital medicine:

essential principles and modern practice. CRC Press.

Maksimov, D., Hesser, J., Brockmann, C., Jochum, S., Di-

etz, T., et al. (2009). Graph-matching based cta. IEEE

Transactions on Medical Imaging, 28(12):1940–1954.

Marcheschi, P., Ciregia, A., Mazzarisi, A., Augiero, G., and

Gori, A. (2009). A new approach to affordable and re-

liable cardiology pacs architecture using open-source

technology. In 2009 36th Annual Computers in Car-

diology Conference (CinC), pages 537–540.

Matthews, B. (2015). Csmd: the core scien-

tiﬁc metadata model. Online http://icatproject-

contrib.github.io/CSMD/csmd-4.0.html.

Mildenberger, P., Eichelberg, M., and Martin, E. (2002).

Introduction to the dicom standard. European Radiol-

ogy, 12(4):920–927.

OASIS (2007). Standard, O.A.S.I.S: Web services busi-

ness process execution language version 2.0. Online

http://docs.oasis-open.org/wsbpel/2.0/OS/wsbpel-v2.

Prabhune, A., Stotzka, R., Jejkal, T., Hartmann, V., Bach,

M., Schmitt, E., Hausmann, M., and Hesser, J. (2015).

An optimized generic client service api for managing

large datasets within a data repository. In Big Data

Computing Service and Applications (BigDataSer-

vice), 2015 IEEE First International Conference on,

pages 44–51.

Prabhune, A., Zweig, A., Stotzka, R., Gertz, M., and

Hesser, J. (2016). Prov2ONE: An Algorithm for

Automatically Constructing ProvONE Provenance

Graphs, pages 204–208. Springer Publishing.

Sanderson, R., Ciccarese, P., Van de Sompel, H., Bradshaw,

S., Brickley, D., a Castro, L. J. G., et al. (2013). Open

annotation data model. W3C community draft.

Sandoval, J. (2009). Restful java web services: Master core

rest concepts and create restful web services in java.

Packt Publishing Ltd.

Zeilinger, G., Montgomery, O., Evans, D., et al. (2010).

The dcm4che project. Online Sourceforge project

https://sourceforge.net/projects/dcm4che/.

Zhang, J., Lu, X., Nie, H., Huang, Z., and van der Aalst,

W. M. P. (2009). Radiology information system: a

workﬂow-based approach. International Journal of

Computer Assisted Radiology and Surgery, 4(5):509–

516.

Zhao, Y., Wilde, M., and Foster, I. (2006). Applying

the Virtual Data Provenance Model, pages 148–161.

Springer Berlin Heidelberg, Berlin, Heidelberg.

Zhu, N. Q. (2013). Data visualization with D3. js cookbook.

Packt Publishing Ltd.

Managing Provenance for Medical Datasets - An Example Case for Documenting the Workﬂow for Image Processing

243