Transcript Grid technology and medical imaging

Grid technology and medical
imaging
Derek Hill
Division of Imaging Sciences
GKT School of Medicine, Guy’s
Hospital
[email protected]
Summary
• Some observations of how computing and imaging
has developed in medical imaging
• What can the grid offer medicine and healthcare
• Two demonstrators:
– Image-based decision support
– Image analysis for drug discovery
N+N meeting 2003
Applications of medical imaging
• Healthcare
–
–
–
–
Patient diagnosis
Patient treatment
Screening
Quality assurance
• Medical research
– Cohort comparisons
– Longitudinal studies
• Drug discovery
– Surrogate end-points in drug trials (pre-clinical and clinical)
• Device development
– Next generation orthopaedic implants
N+N meeting 2003
Image guided interventions
Images Courtesy
Guy’s Hospital
N+N meeting 2003
Image guided interventions II
Images Courtesy
Guy’s Hospital
N+N meeting 2003
Labelling structures using a reference atlas
N+N meeting 2003
Labelling patient images in database
Reference image
(example slice)
Database subject image
(example slice)
N+N meeting 2003
Example labelled subject
Example database subject to whom labelled reference image has
been warped
N+N meeting 2003
Surgical verification
Accuracy of surgical placement against plan
•
•
•
•
Surgeon plans on X-ray or CT, uses database of prostheses
Operation takes place using plan as guidance
Post operative X-ray evaluated for accuracy of placement
Data stored and used for short term assessment and long term evaluation
studies
Courtesy of
Ian Revie
Depuy International
N+N meeting 2003
Support for Multidisciplinary
Collaborative Environments:
Triple Assessment of Breast Cancer
Patients (MIAKTS)
•
•
•
•
•
Surgeons
Radiologists
Pathologists
Oncologists
Nurses
N+N meeting 2003
Multidisciplinary Management
of Breast Cancer
Pathology
Images courtesy of Oxford and Guy’s
Radiology
Surgery
N+N meeting 2003
Magnetic resonance imaging for
breast screening (MARIBS)
• Is MRI an effective way of screening young women
at high risk of breast cancer?
• 17 Centres in the UK (and associated with other
large trials in Europe and Canada)
• Led by the Institute of Cancer Research
• MRC and NHS funded study
N+N meeting 2003
Complex processing for MARIBS and
to support triple assessment
MR Mammogram
Images courtesy of Guy’s Hospital and KCL
Pre-contrast
Subtracted
projection
Post-contrast
Non-rigid
registration
Model deformation
Shape and texture
analysis
N+N meeting 2003
How e-science can help
• E-science is providing:
– An easy-to-use registration service to align and process the
images
– Image-derived metadata that can be queried for clinical
decision support or for research
– Ontologies to improve interoperability of data sources.
N+N meeting 2003
Bone Disease:
What changes do we see in Osteo Arthritis?
1. Joint Space narrowing
2. Changes in Texture
3. Changes in ‘banding’
Courtesy Chris Buckland-Wright
and Lewis Griffin
N+N meeting 2003
Biologists explanations of these changes
involve multiple scales.
Causation flows from fine to coarse
& back down again
N+N meeting 2003
Model at multiple scales
N+N meeting 2003
Distributed computing for mega-scale modelling.
+
-
+
-
+
Fine
Medium Large
Fine
Medium Large
N+N meeting 2003
Size of medical images
• An individual 2D medical image is quite small
–
–
–
–
–
Nuclear medicine: 32kByte
Magnetic Resonance Imaging (MRI): 128kByte
X-ray Computed Tomography (CT): 512kByte
X-ray angiogram: 1Mbyte
Chest x-ray: 16Mbyte
• One patient study is quite large
– Eg: 1 heart study in MRI is typically 1Gbyte
• Aggregated data from cohorts can be very large
– Eg: analysis of 500 subjects
N+N meeting 2003
Image metadata
• Details of image acquisition
–
–
–
–
Modality
Details of acquisition (modality specific)
Geometrical information
Timing information
• Information about patient
– Name, address, doctor’s name, patient identifier
– Past medical history, family history, social history
– Presenting complaint, differential diagnosis
N+N meeting 2003
Characteristics of medical image
analysis software
• Real-time interaction
– Viewing and manipulation of 3D volumes and 2D/3D+time
data
– Interactive structure delineation
• Automatic algorithms
–
–
–
–
Rapid evolution of algorithms (not based on legacy code)
Major area of international research
Algorithm complexity increases faster than Moore’s law
Frequently generate substantial derived information
• Many times the size of the original data
N+N meeting 2003
Medical image storage
• Historically, images have been printed onto film for
storage, and archived in removable media that are
usually unreadable after about 3 years
• Digital medical image archives are becoming
standard (especially in Japan!)
• Patient image storage is distributed (patients often
visit many hospitals over course of their life)
• Many research studies involve multi-site image
acquisition
N+N meeting 2003
Some Observations
• Medical and healthcare industry and hospitals do not regularly
use complex information processing,
– It is not part of their core business to invest in the implementation
and support of this activity
– uptake has been disappointing
• Imaging research and development in academic labs often
stops with the publication of a new method/algorithm
– Yet over the last decade we have seen major advances in many
aspects of this technology (image interpretation, segmentation,
shape analysis, registration, visualisation, ..)
• There is little data sharing except multicentre research studies
where all images send on removable media to central analysis
site.
• International data sharing is problematic
N+N meeting 2003
• Computing is not a core business of healthcare
organisations and related companies (eg: pharma)
• The market is used to paying for services as needed
(eg: image acquisition is paid for on a per-patient
basis, analysis could be the same)
N+N meeting 2003
Potential benefits of the grid
• More effective sharing of data
• More efficient multi-professional working in patient
management
• Access to substantial on-demand computing
resource
• New “collaborations of equals” in which multicentre
studies have full scientific input from all sites
• New ways of image analysis needs being met
– Eg: new companies delivering grid services to healthcare
and pharmaceutical industry.
N+N meeting 2003
Two example applications
• Image-based decision support
• Analysis of images for drug-discovery
N+N meeting 2003
A dynamic brain atlas
Grid-enabled decision support in healthcare
Context
• Better information management is a high priority in
the modernization of the NHS.
• Decision support is a key component
– Existing example: prompting doctor with contra-indications
of selected medicines
• We show how the grid can bring image-based
decision support
– Calculating a customized brain atlas on the fly
N+N meeting 2003
Workflow of busy radiologist
Load patient image from worklist
No
Yes
Easy?
Use
text book
atlas
diagnosis
N+N meeting 2003
Workflow of busy radiologist
Load patient image from worklist
No
Yes
diagnosis
Easy?
Use
patient specific
Dynamic atlas
Viewing
tools
N+N meeting 2003
…need reference data
N+N meeting 2003
200 reference subjects
Example slices
From MRI
Volume
images
N+N meeting 2003
IMPERIAL
COLLEGE
Get reference images
KING’S COLLEGE
King’s
College
LONDON
London
(Guy’s Campus)
Patient scan
+ instructions
Oxford
University
N+N meeting 2003
IMPERIAL
COLLEGE
KING’S COLLEGE
King’s
College
LONDON
London
(Guy’s Campus)
Oxford
University
N+N meeting 2003
KING’S COLLEGE
King’s
College
LONDON
IMPERIAL
COLLEGE
London
(Guy’s
Campus)
Create
atlas
atlas
Oxford
University
N+N meeting 2003
The Radiologist’s view
N+N meeting 2003
N+N meeting 2003
N+N meeting 2003
N+N meeting 2003
Conclusions
• The dynamic atlas provides a customized
authoritative reference presented in an intuitive way
• The doctor can see at a glance the normal range of
sizes and shapes of each brain structure, overlaid on
the patient’s own scan, assisting diagnosis.
• The grid will bring new ways of working to Healtcare
N+N meeting 2003
Team
• Derek Hill, Thomas Harkens, Kate McLeish, Colin
Renshaw, King’s College London (Guy’s Campus)
[email protected]
• Jo Hajnal, Imperial College London (Hammersmith)
[email protected]
• Daniel Rueckert, Imperial College London (South
Ken) [email protected]
• Steve Smith, University of Oxford
[email protected]
N+N meeting 2003
Grid services in the drug-discovery
workflow
Context
• Pharmaceutical companies are major users of
imaging
• They need validated automated image analysis to
quantify drug efficacy for surrogate endpoints
N+N meeting 2003
Drug discovery
Scientist
The Grid
Bone labelling service,
brain labelling service, …
N+N meeting 2003
Demonstrator system
3. Registration job
running
1. Locate image data
2. Transfer data
by ftp or grid-ftp*
IXI
GSK
4. Download results
Image registration service
*think of grid-ftp as a secure version of ftp
N+N meeting 2003
Commercial opportunities
• Specialist companies will provide complex
information processing services (eg in medical image
analysis)
• They will purchase computing resource as needed
• Their customers will be:
–
–
–
–
Hospitals, PCTs
The pharmaceutical industry
Medical devices industry
Government agencies
N+N meeting 2003
Conclusions
• Medical imaging is well suited to grid capabilities
• There are particular problems of security and
confidentiality
• There is less legacy s/w and hardware in medical
imaging than in some other scientific and
engineering applications
N+N meeting 2003
Thankyou
N+N meeting 2003