Leukemia cell classifier — live demo
Select a real single-cell image below. The v3 model classifies it as leukemic (a B-ALL blast) or a normal cell, in your browser. The samples come from three different labs — Delhi, Tehran and Barcelona — to show it works across sources. Nothing is uploaded.
- Diagnosing leukemia requires a hematologist, bone-marrow analysis, flow cytometry, and genetic testing — never a single cell image.
- This is a demonstration of image classification on research datasets. Even the honest, multi-lab model is imperfect (the write-up reports where it false-alarms) — treat every output as an educational example.
- If you have any health concern, consult a qualified clinician.
Select a cell
Each thumbnail is a real single-cell image, drawn from three separate labs (Delhi, Tehran, Barcelona) — hover to see each cell's source — so you can watch the model classify cells from microscopes it has to generalize across.
This is the multi-source v3 model — it generalizes across labs far better than the single-source version, but it is still imperfect. The write-up reports the honest numbers, including where it false-alarms on normal cells.
What am I looking at?
Each image is a single white blood cell, stained and photographed under a microscope. A leukemic cell is a B-ALL lymphoblast; a normal cell is a healthy one — including normal lymphocytes, which look deceptively similar. Distinguishing them is subtle even for specialists, and doing it across different labs' microscopes is harder still — which is the whole point of v3.