Pulse.

a daily field guide to health research that matters

◆ Console

‹ Sat · 16 May 2026
Near-term implementable finding

Multicentric data challenge for artificial intelligence-based classification of leukocytes: results from the CytologIA consortium

AI models trained on 69,000 blood cell images now classify white blood cells nearly as accurately as human experts, potentially speeding up diagnoses.

The CytologIA consortium organized the first large-scale, open, multicentric AI benchmark for hematological leukocyte morphology classification, with 245 teams competing on 69,168 annotated images spanning 23 leukocyte classes. The top model achieved balanced accuracy of 0.94, establishing a new reproducibility reference for AI-assisted hematological diagnostics.

What the study was

Study design
Multicentric AI benchmark challenge with expert-annotated open dataset
Population
20 hematology laboratories (France, Belgium, Switzerland); 69,168 peripheral blood smear images; 245 participating teams
Sample size
69168
Category
Diagnostics
Maturity
Validated
Journal
NPJ Precis Oncol

Why it surfaced

First large-scale, open, multicentric AI benchmark for leukocyte morphology classification. Published in NPJ Precision Oncology. Open data release enables rapid clinical adoption. Directly addresses variability in manual hematological morphology assessment. Highest score in CBC/ML topic this run.

A plain-language summary of published research — not medical advice. Talk to a clinician about your own care.