An open bone marrow megakaryocyte dataset for automated morphologic studies.

Area of Science:

• Hematology
• Computational Biology
• Medical Imaging

Background:

• Accurate megakaryocyte classification is vital for diagnosing hematological disorders such as Myelodysplastic Syndromes (MDS).
• Existing deep learning (DL) applications are limited by a lack of high-quality, openly licensed megakaryocyte datasets.
• This gap hinders the development of automated diagnostic tools.

Purpose of the Study:

• To introduce MK-11, the first publicly available dataset for megakaryocyte subtype classification.
• To establish a benchmark for evaluating DL models in megakaryocyte morphological assessment.
• To facilitate research in hematological disorders and platelet production.

Main Methods:

• Curated a dataset (MK-11) of 7,204 Wright-Giemsa stained single-cell images covering 11 megakaryocyte subtypes.
• Images were annotated by two experienced hematopathologists using standardized criteria and consensus review.
• Evaluated state-of-the-art deep learning models (CNNs, Transformers) on the dataset.

Main Results:

• Established strong baseline performance for megakaryocyte classification using various neural networks.
• Demonstrated the feasibility of DL models for automated megakaryocyte subtype identification.
• Provided standardized cross-validation partitions and evaluation scripts for reproducibility.

Conclusions:

• MK-11 is the first public dataset enabling the development and evaluation of automatic megakaryocyte morphological assessment.
• This resource will accelerate research and diagnostic tool development for MDS and related platelet disorders.
• The dataset and associated code are released under open licenses to promote further research.