GRIDs for biomedical applications

1
GRIDs for biomedical applications

• FP6 project meeting, 5.7.2005

Henning Müller Service of Medical
Informatics Geneva University Hospitals
2
Overview

• University Hospitals of Geneva
• Medical Informatics Service
• Applications for the GRID
• Biomedical text mining
• Medical image indexing retrieval
• Role and conditions

3
University Hospitals of Geneva

• 2,200 beds, 6 hospitals
• 9,000 employees
• 1,300 MDs
• 6,000 computers (windows)
• Central administration anddistribution of
  software
• Infrequent use of most computers
• Budget of hospital 1 billion/year
• 1,200 medical students
• 3,000 computers in medical faculty building
• Bioinformatics, Library, Labs,

4
Service of medical informatics

• 60 employees, part of the radiology department
• 10 persons in research
• Research areas
• Multimedia electronic patient record
• Decision support systems
• Telemedicine, especially with African countries
• Knowledge representation, natural language
  processing, data mining
• Image processing, PACS, operation planning
• Teaching
• Postgraduate course in medical informatics
• Virtual campus for medical students in medical
  informatics

5
Application overview

• Large amount of data produced in hospital
• 780,000 patient days per year
• 30,000 images produced per day
• Data warehouse available for analysis of old
  datasets for strategic planning
• No automatic analysis of data sets, only on
  demand
• Applications concerning the analysis and
  retrieval of visual and textual information
• Biomedical text mining
• FP6 NoE SemanticMining
• Content-based image retrieval
• Open Source framework GIFT (GNU Image Finding
  Tool)
• Worldwide expert in medical image retrieval
• 100 publications, gt20 this year

6
Planned architecture
Web/Application server
ARC front-end
Computing back-end (Windows desktops running BOINC
)
Storage
GUI for users
7
Application biomedical text mining

• Texts are abstracts of publications from Medline
  (Genomics TREC)
• 5 million abstracts of ten years (rising
  strongly), 20 GB text in XML format
• Full-text analysis is foreseen for smaller set
  (meaning more data)
• Goal Fulfill information needs of a user with
  respect to articles concerning certain genes or
  sequences
• Treatment is increasingly complex to extract
  meaningful information
• Analysis of medical texts from the electronic
  patient record
• Millions of entries each year
• Not systematically analysed at the moment

8
Application content-based image retrieval

• Goals to better manage the large amount of
  images produced (gt30.000 per day)
• Automatic control of DICOM header information
  with visual data
• Retrieval of similar visual images from
  restricted collections, mainly for teaching
  (50.000 images)
• Visual analysis images by image to generate
  visual and textual index
• Retrieval is quick and can be parallelised
• System works on linux, but feature extraction
  could be transformed for Windows
• Diagnostic aid in restricted domains by supplying
  similar cases to an example with images
• Lung CT analysis
• Dermatology

9
Image retrieval example
10
Role and conditions

• User of grid services and tests of a grid
  infrastructure
• Adaptation of applications to use on a grid
• Tests of infrastructure with these applications
• Tasks 1 PhD student at 80
• Supervision University of Geneva medical and
  sciences, CERN
• Background study of technologies, getting
  certificates ( 4 months)
• Setting up BOINC, ARC ( 4 months)
• Application programming ( 9 months)
• Prototype launch ( 3 month)
• Tutorials for users, incorporating changes if new
  user requirements ( 3 months)
• Production launch ( 3 month)
• Report writing ( 3 months)

11
Conclusions

• Grid technology is a strategic initiative for the
  University and Hospitals (long term)
• Goal better use the existing infrastructure
• No financing for high-performance computing is
  available
• Explore possibilities to use a service provider
  for important tasks if needed (to gain
  flexibility and speed)
• Large amount of data produced and often not used
  in a systematic way
• Applications are well adapted to be griddified
• More applications than the stated ones seem well
  possible
• Simulations of blood flow,