République
Française
CytologIA
1. Purpose of the database:
The aim of the Data Challenge is to develop an artificial intelligence capable of correctly classifying images of physiological or pathological blood leukocytes in order to provide the most accurate haematological diagnosis possible. To achieve this, 19 French-speaking centres were mobilised via the GFHC to build up the database, which is made up of 23 different classes, thus representing the most varied database of leukocyte images in the world.
2. Context of creation of the database:
The database was created following a call for projects for the Health Data Hub's ‘Data Challenge en Santé’.
3. Target:
Haematologists, biologists, data scientists specialising in image recognition, pathologists
4. Results obtained from the database:
245 participants from all over the world took part in the Data Challenge. The 3 best algorithms showed performance scores of between 0.93 and 0.94, which is excellent and rarely achieved in multricentre leucocyte recognition studies.
5. Other informations:
- Collection methodology and inclusion criteria:
Each centre sends the leukocyte images to the project leader, who centralises the collection. The images are classified by leukocyte type and, within the group of pathological leukocytes, these cells are classified according to haemopathy. The main inclusion criterion is to have a patient with haemopathy and to send images of sufficient quality, with individualised cells on the photo. These images must be anonymous. Approximately 70,000 images classified by leukocyte type (23 classes) with at least one thousand images per class
- Choice of variables:
Anonymity, image format (JPEG, PNG, TIFF), physiological or pathological leukocyte on a blood smear.
- Support:
6. Licences:
Creative Commons Attribution (CC BY 3.0) Licence Ouverte/Open Licence 2.0 (Etalab 2.0)
6. Cite:
For any reuse of this database, use the DOI provided: http://doi.org/10.60597/9st8-xe88
- Dernière mise à jour
- 3 juin 2025