medGIFT ContentBased Access to Medical Images

1
medGIFT Content-Based Access to Medical Images
Henning Müller Division of Medical
Informatics University Hospitals of
Geneva 19.6.2003
2
Content-based access to visual data

• Access based only on visual features
• Grey levels (local, global, distribution)
• Textures
• Forms of segments
• Show me images that are similar to
• No semantics in features, only complement to text

Emphysema
Emphysema
?
Micro nodules
Macro nodules
3
The medGIFT project

• Uses an open source tool http//www.gnu.org/softwa
  re/gift/ developed in the Viper project
• Adaptations of the features for medical purposes
• Integration into CasImage is in process
• Indexation of the images
• Demo at http//cih-dm26.hcuge.ch/demo/demo.php
• Communication interfaces with Casimage for
  complete integration
• Specialization for lung images is planned
• Possible benefits currently for teaching and
  research
• Long-term goal is diagnostic aid

4
Future projects with medGIFT

• Demand Fonds National for the project
• Specialization, integration, improvement of the
  project
• Medical study on use and acceptance of image
  retrieval
• Combination of textual and visual characteristics
• Quality control
• Semi-automatic annotation
• Improve the retrieval quality
• Integration into the PACS
• Very long-term as the amount of data is huge

5
Conclusions

• Large production of visual content in medical
  departments
• Alternative tools will be needed to manage these
  data
• Multimedia data mining
• Retrieval using various sources
• Content-based visual data access has a lot of
  potential
• Available programs need to be adapted for the use
  in medical applications and specialized for
  certain domains
• So far mainly research prototypes
• Comments or propositions are always welcome