medGIFT retrieving medical images based on their visual content

1
medGIFT retrieving medical images based on
their visual content
Henning Müller Service of medical
informatics University hospitals of
Geneva Geneva, Switzerland

• SIS-Forum, ICT4D, WSIS
• Geneva, 9.-13.12.2003

2
What will be presented

• The University Hospitals of Geneva
• Service of medical informatics
• The role of open source
• Content-based visual data access
• Medical image retrieval
• GIFT and medGIFT
• Benefits from visual data access
• demonstration

3
The University Hospitals of Geneva

• 2,200 beds in 6 hospitals
• Radiology gt13,000 images produced a day
• Medical Informatics 60 people, many projects
• Service for the hospital
• Electronic patient record
• Picture Archive
• Research
• Image processing, Natural Language Processing
• E-toile
• Teaching
• Medical students, postgraduate course
• Telemedicine

4
The role of Open Source Software

• Source Code is freely available
• Better software quality
• Better documentation of source code
• Vendor independence (no out of business problems)
• Large base of potential developers
• Publicly available knowledge
• Software is often free of charge
• The service needs to be paid
• In developing countries often the only option
• Publicly funded research projects should be OS

5
Content-based visual data access Situation
6
Content-based visual data access

• Developed due to the exploding creation of
  multimedia data and the availability of it often
  on the web, but also within companies (CNN)
• Annotation is expensive, thus only possible in
  commercially interesting domains
• Subjectivity, spelling errors,
• Goal is to have a data access based on
  automatically extracted visual features
• Query by example Show me images like this one
• Problem to have a starting point for a query
• Advantages and problems need to be understood for
  proper use

7
Content-based visual data access example
Blobworld, Berkeley
8
Medical image retrieval

• Access currently mainly by patient ID
• No use of implicitly stored knowledge in the
  images
• With film, nothing else is possible
• Teaching databases offer text and structured data
• Possibilities of content-based access
• Teaching
• Lecturers
• Students
• Research
• Diagnostic aid
• Especially for less experienced professionals

9
An example
Emphysema
Micro nodules
?
Emphysema
Macro nodules
10
GIFT and medGIFT

• GNU Image Finding Tool (GIFT)
• Framework of components
• Server, clients, databases, feature extractors,
  query language
• Open Source
• medGIFT
• Medical adaptation of GIFT
• More importance for gray scales and textures
• Interface that shows the diagnosis
• Link with the teaching file
• Specializations are foreseen
• Research project

11
The Interface
12
Conclusions

• Hospitals need science/research to advance in
  teaching and patient care
• Open Source should be model for publicly funded
  research projects to create publicly available
  knowledge, and to reuse software
• We will need new tools to manage the large
  amounts of audiovisual data that are being
  created
• Content-based visual data access is one such tool
• The introduction of it in the medical domain
  needs an effort of all partners and can pay off
  in the long run