Blood Cell Counter
Automatic cell counts for bird blood smears
Blood Cell Counter
Upload blood smear slides — counts erythrocytes and leukocytes automatically.
Advanced settings
Automatic treats .jpg/.png as microscope photos and .svs/.ndpi as scans.
Photos are shrunk to the scale the model learned from. 8 suits a 100× oil objective; about 3 for 40×.
How sure the model must be before a cell counts. Default 0.5.
How strict it is about which parts of the smear are good enough to count (monolayer). Default 0.7.
The same field of view, side by side: the original Marburg model, the model in use, and your check when it is a photo you have checked. Click a picture to enlarge it.
Loading…
Photos uploaded with the same sample name, added up. A photo you checked counts with your numbers. Tap a sample for its full counts.
Loading…
Loading…
Nothing trains on its own: only when you press Train now, or at the times you set here. How training works
The cells where your marks and the model in use disagree, the ones it is surest of first. Some are the model's mistakes, some may be yours. Tap the right type to change your mark, or Keep mine if yours is right; either way the cell isn't shown again. Change a mark only when you are sure: the model's view is no evidence on its own, and on photos it never learned from, its later test score would otherwise lean its way. Keys on a selected cell: h l e m b t r, x not a cell, k keep mine, o open the photo.
Loading…
Counting uses AviBloodCount (Vogelbacher et al., Birds 5, 2024). The models were trained on whole-slide scans at 40×, so microscope photos are first shrunk to that scale (8× for a 100× oil objective — change it under Advanced settings for other objectives) and then counted as a whole. Slide scans are cut into 512×384 px tiles and only the tiles where cells lie in a countable monolayer are counted. The training slides were stained pink-purple; on differently stained smears, check the detections, especially the leukocyte types. Parasitised erythrocytes are detected but not reported, as in the original software. Once a model trained on your checked photos is in use (see the Model tab), it finds and types the leukocytes in microscope photos instead.