Related Experiment Videos
A reference data set for the evaluation of medical image retrieval systems
Henning Müller1, Antoine Rosset, Jean-Paul Vallée
1University Hospitals of Geneva, Service for Medical Informatics, Rue Micheli-du-Crest 24, 141211 Geneva, Switzerland. henning.mueller@sim.hcuge.ch
Summary
A new, large, and freely available medical image database has been created for research. This database aids in comparing content-based image retrieval systems and supports medical education.
Area of Science:
- Medical Imaging
- Information Retrieval
- Computer Science
Background:
- Content-based image retrieval (CBIR) is crucial for medical imaging research and management.
- Existing CBIR systems face challenges in performance comparison due to proprietary/private image databases and small dataset sizes.
- Lack of standardized, accessible databases hinders evaluation and advancement in the field.
Purpose of the Study:
- To describe the creation of a large, freely accessible medical image database.
- To provide a resource for evaluating and comparing CBIR systems in medical contexts.
- To support medical education and teaching file development.
Main Methods:
- Compilation of over 8,700 anonymized medical images for a teaching file.
- Generation of ground truth for 26 query images using pooling methods from information retrieval.
- Ensuring images are free of copyright and patient privacy concerns.
Main Results:
- A substantial, anonymized medical image database is now available for research.
- Ground truth data was generated for a subset of images to facilitate evaluation.
- The database serves as a valuable starting point for comparing CBIR systems.
Conclusions:
- The developed database addresses the need for accessible resources in medical image retrieval research.
- It facilitates the comparison of retrieval systems, particularly for educational purposes.
- Further development of specialized databases with diagnostic information is needed for clinical applications.