Motivation
The project
The middleware
Applications
Summary
GIMI:
Generic Infrastructure for Medical Informatics
Andrew Simpson, on behalf of the GIMI consortium
Oxford University Computing Laboratory
July 2008
Simpson et al
GIMI
Motivation
The project
The middleware
Applications
Summary
1
Motivation
2
The project
3
The middleware
sif
Plug-ins and data agnosticism
Evolving access control
4
Applications
Long-term conditions
Mammography auditing and training
Medical imaging in cancer care
Other applications
5
Summary
Simpson et al
GIMI
Motivation
The project
The middleware
Applications
Summary
Motivation
Medical (and other) researchers often have significant
amounts of data collected over a period of years that they
would like to provide access to in real-time: to improve
patient care; to act as a training resource; to achieve
‘bigger and better research’; to ‘realise assets’
On the other side of the equation, such researchers often
wish to ‘get at’ large data sources in real-time
The facilitation of secure data access, sharing and
integration across organisational boundaries is essential in
both cases
These issues are, of course, also relevant in many other
contexts
Simpson et al
GIMI
Motivation
The project
The middleware
Applications
Summary
Motivation
“Electronic records could facilitate new interfaces between
care and research environments, leading to great
improvements in the scope and efficiency of research.
Benefits range from systematically generating hypotheses
for research to undertaking entire studies based only on
electronic record data . . . Clinicians and patients must have
confidence in the consent, confidentiality and security
arrangements for the uses of secondary data. Provided that
such initiatives establish adequate information governance
arrangements, within a clear ethical framework, innovative
clinical research should flourish. Major benefits to patient
care could ensue given sufficient development of the
care-research interface via electronic records.”
J. Powell and I. Buchan.
Electronic health records should support clinical research. Journal of Medical Internet Research, 7(1):e4, 2005.
Simpson et al
GIMI
Motivation
The project
The middleware
Applications
Summary
Project aims
The main aim of GIMI is to develop a generic, dependable
middleware layer capable of:
1
(in the short term) supporting data sharing across
disparate sources to facilitate healthcare research,
delivery, and training;
2
(in the medium term) facilitating data access via dynamic,
fine-grained access control mechanisms
3
(in the longer-term) interfacing with technological solutions
deployed within the NHS
The technology development is being validated via three
applications
Simpson et al
GIMI
Motivation
The project
The middleware
Applications
Summary
Partners
University of Oxford (Computing Laboratory and
Engineering Science)
University College London
Loughborough University
t+ Medical
IBM UK
Siemens Molecular Imaging
The National Cancer Research Institute
Funding from TSB; total project value 2.4M pounds
Simpson et al
GIMI
Motivation
The project
The middleware
Applications
Summary
Work-packages
Work-package 1: project management (OUCL)
Work-package 2: core technology (OUCL)
Work-package 3: long-term conditions (t+ Medical and
OUES)
Work-package 4: mammography auditing and training
(UCL and Loughborough University)
Work-package 5: medical imaging in cancer care (OUES)
Work-package 6: clinical data management and
interoperability (OUCL)
The project structure corresponds to a ‘hub-and-spoke’ model
Simpson et al
GIMI
Motivation
The project
The middleware
Applications
Summary
sif
Plug-ins and data agnosticism
Evolving access control
The middleware
The aims of the project give rise to two distinct—but
complementary—technologies:
sif (service-oriented interoperability framework), and
evolving access control
Simpson et al
GIMI
Motivation
The project
The middleware
Applications
Summary
sif
Plug-ins and data agnosticism
Evolving access control
Not another middleware framework . . .
Yes . . . but . . .
Previous experience (e-DiaMoND—based on a
combination of GT3, OGSA-DAI and IBM products) and a
lack of options three or so years ago (WSRF panned by
Watson and Parastatidis; GT4 not yet released; OMII-UK
not particularly portable at that point) meant that going our
own way and using a WS-I+ (with our + being
WS-Security) approach was the best option for us
Data is at the heart of our concerns, so we wanted a
data-oriented solution . . .
and we also wanted a portable and interoperable one
Simpson et al
GIMI
Motivation
The project
The middleware
Applications
Summary
sif
Plug-ins and data agnosticism
Evolving access control
Abstraction
Issues pertaining to secure transfer and data federation
are abstracted from the end user
The middleware is agnostic to the kind of data that is
shared; furthermore, it is agnostic to what is done to that
data
Via a ‘plug-in’ mechanism, domain specialists develop
applications to access and manipulate remotely held data
We aim to facilitate technology-agnosticism via
data-agnosticism
Simpson et al
GIMI
Motivation
The project
The middleware
Applications
Summary
sif
Plug-ins and data agnosticism
Evolving access control
Interoperability
We assume that legacy systems have integral value: there
are no pre-conceived schemas, ontologies or interfaces
that render legacy data incompatible
It is the applications and communities that determine
compatibility
It is our view that lower level interoperability should be
achieved via the use of open standards; higher level—or
semantic—interoperability should be achieved on an
application-by-application, or domain-by-domain basis
The aim is to facilitate ‘bottom-up’—rather than
‘top-down’—virtual organisations
Simpson et al
GIMI
Motivation
The project
The middleware
Applications
Summary
sif
Plug-ins and data agnosticism
Evolving access control
sif and GIMI
VO1
VO3
VO2
App 1
App 2
App 3
sif
Source 1
Source 2
sif
Source 3
Simpson et al
Source 4
GIMI
Source 5
Motivation
The project
The middleware
Applications
Summary
sif
Plug-ins and data agnosticism
Evolving access control
Secure data sharing
The responsibility for determining access to data resides
with the data’s owner: it is not for us to impose
pre-conceived policies on data owners; as such,
expressive and flexible models should be in place to
accommodate users’ needs
We are interested in abstract—but not
restrictive—representations
We are also interested in utilising audit information to
facilitate the construction of ‘meta-policies’
Simpson et al
GIMI
Motivation
The project
The middleware
Applications
Summary
sif
Plug-ins and data agnosticism
Evolving access control
Technology drivers
The use of technologies that are:
portable,
interoperable,
standards-based,
freely available, and
(where possible) open source
resulting in middleware that:
meets our requirements with respect to abstraction,
interoperability and data ownership (amongst others),
requires minimum buy-in from data owners, and
is straightforward for application developers to pick up and
play with
Simpson et al
GIMI
Motivation
The project
The middleware
Applications
Summary
sif
Plug-ins and data agnosticism
Evolving access control
A health grid architecture
Virtual Organisation
E
P1
I
E
P2
Data
Data
ws1
I
ws2
Hospital 1
Simpson et al
Hospital 2
GIMI
Motivation
The project
The middleware
Applications
Summary
sif
Plug-ins and data agnosticism
Evolving access control
Federated database access
Simpson et al
GIMI
Motivation
The project
The middleware
Applications
Summary
sif
Plug-ins and data agnosticism
Evolving access control
Plug-ins
We wish to offer functionality for three types of plug-in:
data, file, and algorithm
By using a standard plug-in interface for each of the three
types it becomes possible to add heterogeneous resources
into a virtual organisation
There is no need for the resource being advertised through
the plug-in system to directly represent the physical
resource—what is advertised as a single data source may
come from any number of physical resources, or even
another distributed system
Simpson et al
GIMI
Motivation
The project
The middleware
Applications
Summary
sif
Plug-ins and data agnosticism
Evolving access control
Plug-ins (continued)
It is envisaged that the following will be ultimately accessible via
plug-ins:
Relational databases
XML databases
Object databases
Picture Archiving and Communication Systems (PACS)
Local file systems
Network file systems
GIMI file system
NHS Secondary Uses Service
Algorithms
Simpson et al
GIMI
Motivation
The project
The middleware
Applications
Summary
sif
Plug-ins and data agnosticism
Evolving access control
Plug-in authoring
Plug-in types have different schemas to conform to
Each schema allows the author to define the interface to
their plug-in
The plug-ins are implemented in Java and packaged as Jar
files
Can support embedded libraries and supporting files
A plug-in defines any translation necessary between
common plug-in interfaces and actual back-end
implementation
Simpson et al
GIMI
Motivation
The project
The middleware
Applications
Summary
sif
Plug-ins and data agnosticism
Evolving access control
Evolving access control
Via XACML we offer the potential for fine-grained access
control policies
Furthermore, by coupling the access and audit
mechanisms and allowing the description of meta-policies,
data access can change as a consequence of observed
actions
The approach permits the expression of meta-policies
(policies about policies), which allow policies to evolve on
the basis of observed events
Relevant events are observed as all external data access
is channelled through a single access point
Simpson et al
GIMI
Motivation
The project
The middleware
Applications
Summary
Long-term conditions
Mammography auditing and training
Medical imaging in cancer care
Other applications
Long-term conditions
There is a drive towards self-management as a means of
improving the health of patients with long-term conditions
This needs to be supported by comprehensive IT systems
for disease management, which integrates all relevant
information
Initial research has focused on the development of robust
algorithms for alerting healthcare professionals when a
patient’s data deviate from the expected pattern
Telemedicine trial involving two GP surgeries from the local
PCT—involving delivering blood-glucose summary data
from a telemedicine handset to a GP’s desktop
Simpson et al
GIMI
Motivation
The project
The middleware
Applications
Summary
Long-term conditions
Mammography auditing and training
Medical imaging in cancer care
Other applications
Application 1 authorisation requirements
One application allows a patient to use their mobile phone
to send pertinent information to a central server, where it is
recorded; second allows GPs to access the data pertaining
to their own patients
This gives rise to the requirement that a particular doctor
can only view data for their own patients
t+ Medical are providing additional applications that make
use of the plug-in mechanism to provide processed data
Access to the various plug-ins also requires restricting to
only those who are permitted access to the processed data
Simpson et al
GIMI
Motivation
The project
The middleware
Applications
Summary
Long-term conditions
Mammography auditing and training
Medical imaging in cancer care
Other applications
Mammography auditing and training
The main aim is to develop a prototype training tool for
screening mammography which could offer radiologists a
unique educational experience based on:
intelligent selection of training activities,
a large number of richly annotated images (previous
PERFORMS cases),
a dataset annotated with specific learning goals, and
e-learning materials
Simpson et al
GIMI
Motivation
The project
The middleware
Applications
Summary
Long-term conditions
Mammography auditing and training
Medical imaging in cancer care
Other applications
Application 2 authorisation requirements
The application requires access to health-related data,
together with data associated with the person undergoing
the training or assessment
“Only the person who the data relates to and their
supervisor (or those approved by the data subject) can see
the data”
Institutions have indicated that they need to be able to
control who has access to any given annotated training
data set
It is important to be able to define policies that depend on
the actions being performed: a radiologist undergoing
assessment should not have any cases from their own
institution contained within the data comprising the test set
Simpson et al
GIMI
Motivation
The project
The middleware
Applications
Summary
Long-term conditions
Mammography auditing and training
Medical imaging in cancer care
Other applications
Medical imaging in cancer care
The overall aim of this application area is to:
develop medical image algorithms for application to breast
and colorectal cancer,
build on previous projects, notably eDiaMoND,
Mammogrid, and the NCRI Informatics Initiative
demonstrator on colorectal cancer, and to
provide a testbed and user feedback for prototypes as they
are developed throughout the duration of the project
The application is primarily concerned with providing diagnostic
information based on image analysis of mammograms; to
validate these algorithms, researchers require access to a large
volume of image data from research archives and live
healthcare systems
Simpson et al
GIMI
Motivation
The project
The middleware
Applications
Summary
Long-term conditions
Mammography auditing and training
Medical imaging in cancer care
Other applications
Application 3 authorisation requirements
“Only allow researcher Y to see anonymised data
pertaining to the following group of patients”
“No researcher may see any data pertaining to the
following patients”
“Only data pertaining to the following patients may be
viewed”
“Each researcher has access to a customised list of
attributes pertaining to particular patients”
“A researcher can only see patients involved in a particular
trial”
Simpson et al
GIMI
Motivation
The project
The middleware
Applications
Summary
Long-term conditions
Mammography auditing and training
Medical imaging in cancer care
Other applications
Other applications
NeuroGrid
A prototype demonstrator for the NCRI
Facilitating access to a Dementia data set based at the
John Radcliffe Hospital in Oxford
Simpson et al
GIMI
Motivation
The project
The middleware
Applications
Summary
Simpson et al
Long-term conditions
Mammography auditing and training
Medical imaging in cancer care
Other applications
GIMI
Motivation
The project
The middleware
Applications
Summary
Simpson et al
Long-term conditions
Mammography auditing and training
Medical imaging in cancer care
Other applications
GIMI
Motivation
The project
The middleware
Applications
Summary
It’s all about the data . . .
GIMI has been driven by the needs of appropriate, secure
data sharing
Rather than adopt an existing toolkit-based approach, a
lightweight system built from Java and web services (with a
bit of C that is hidden away) has been developed
The need for fine-grained, expressive, user-defined
authorisation policies, coupled with strong auditing
mechanisms—as well as the requirement that the tasks of
sharing data should be simplified as much as
possible—has led to the adopted approach
Simpson et al
GIMI
Motivation
The project
The middleware
Applications
Summary
What the middleware doesn’t do
Data ‘grids’, not compute grids
Semantic interoperability
Data management
Simpson et al
GIMI
Motivation
The project
The middleware
Applications
Summary
Going on our own
Amongst other things we have had to implement the following
things that would have come as standard using a grid toolkit:
Certificate Authority(CA) management
Use openssl
Proxy certificates
Have a custom ticketing system
Service registries
Use a peer-to-peer system to locate nodes
Every node has the same set of services
File storage and retrieval
Each node has a simple file system
Files are stored and retrieved from WebDAV folders
Remote code execution
Computation is performed using algorithm plug-ins which
typically run on their own servers
Simpson et al
GIMI
Motivation
The project
The middleware
Applications
Summary
And there’s still ‘lashing’ together to do . . .
It’s just the concerns are slightly different (“will Sun’s extensions
to the WS-security libraries work with IBM JVMs?”)—but we are
now where we wanted to be three years ago—and, most
importantly, are now in a position to build applications for new
collaborators in a matter of weeks
Simpson et al
GIMI
Motivation
The project
The middleware
Applications
Summary
Progress
Simpson et al
GIMI