AI cytometry publications

ESCCA 2026 poster. A real-world multicenter evaluation of AI-assisted leukocyte population identification, run with the Laboratorio Especializado de Hematología at Hospital Dr. Rafael Ángel Calderón Guardia (CCSS) in Costa Rica, the HpH Institut für Pathologie und Hämatopathologie Hamburg and Pathodiagnostik Berlin MVZ GmbH.

The study asks whether one AI model, trained predominantly on Western European cohorts, keeps its accuracy across different laboratories, panels, workflows, sample types and patient populations. The same model was applied retrospectively at all three sites, with no site-specific retraining and no panel-specific optimization.

• 7,039 samples across three laboratories in two countries
• Excellent concordance for 8 of 9 shared populations, with Pearson ρ from 0.93 to 1.00
• Pooled linear-fit slopes from 0.93 to 1.13, near unity
• High agreement against both routine clinical differential counts and expert manual gating
• Monocyte identification was the only population with reduced agreement

View the full poster or download it as a PDF.

The AI model is hema.to CellStudio (RUO): a proprietary transformer-based neural network trained from scratch on thousands of expert-annotated flow-cytometry files. Research use only; not for use in diagnostic procedures.

ESMO AI 2025 poster. This work on AI cytometry was led by hema.to’s partners at the HpH Institut für Pathologie und Hämatopathologie Hamburg and Pathodiagnostik Berlin MVZ GmbH.

The retrospective comparison across more than 6,500 peripheral-blood samples shows high concordance with conventional laboratory results across major subpopulations, robustness across multiple panels, workflows, and laboratory locations, and where further harmonization is needed.

Download the full poster.

Clinical study white paper. hema.to was clinically validated in an international, four-center clinical trial. Blood cancer diagnoses were made twice as fast, with 8% more sensitivity and 10% more specificity, using hema.to’s decision-support software.

Download the white paper.

Cell Patterns paper. Nanditha Mallesh and co-authors describe how knowledge transfer enhances AI performance in Cell Patterns.

The collaboration between hema.to’s team, the Munich Leukemia Lab, Charité, and additional university clinics showed that data from multiple laboratories can be combined to increase the clinical performance of AI models for each laboratory.

Cytometry A paper. hema.to’s team collaborated with experienced hematologists and the University of Bonn on this Cytometry A study.

Zhao et al. showed that deep learning can aid routine clinical workflows using cytometry data.

Blood paper II. Nanditha Mallesh and co-authors presented this 2019 Blood paper.

The study showed that combining multiple datasets improves diagnostic performance through an approach called “knowledge transfer.”

Blood paper I. A collaboration between hema.to, the Munich Leukemia Lab, and the University of Bonn produced this 2019 Blood paper.

The paper shows that AI can support hematologists in routine blood cancer diagnosis workflows by providing highly reliable recommendations.