Curated multi-source dataset for AI-based peripheral blood cell classification and computer-aided diagnosis with a unified 14-class taxonomy

Curated multi-source dataset for AI-based peripheral blood cell classification and computer-assisted diagnosis with an unified 14-class taxonomy This dataset is a curated collection of single-cell images of peripheral blood cells, designed for machine learning and artificial intelligence applications in hematological image analysis and computer-aided diagnosis. The dataset integrates three publicly available sources to ensure variability in staining protocols, imaging conditions, and annotation granularity: - AML-Cytomorphology dataset (Matek et al.) - Raabin-WBC dataset (Kouzehkanan et al.) - Multi-focus WBC dataset (Park et al.) All images correspond to cropped single-cell regions extracted from peripheral blood smears and annotated by expert hematopathologists or trained specialists. The primary contribution of this dataset lies in the harmonization of heterogeneous annotation schemes into a unified 14-class taxonomy that preserves biologically meaningful relationships between cell types and their developmental stages. Source Datasets 1. AML-Cytomorphology Dataset Matek et al. (2019) provide a dataset of single-cell images from peripheral blood smears of patients with acute myeloid leukemia and non-malignant controls. Reference: Matek, C., Schwarz, S., Spiekermann, K., & Marr, C. (2019). Human-level recognition of blast cells in acute myeloid leukemia with convolutional neural networks. *Nature Machine Intelligence*, 1(11), 538–544. From this dataset, the following classes were retained: - Basophil - Eosinophil - Erythroblast - Lymphocyte (typical) - Atypical lymphocyte - Monocyte - Myelocyte - Metamyelocyte - Band neutrophil - Neutrophil (segmented) - Promyelocyte - Myeloblast 2. Raabin-WBC Dataset Kouzehkanan et al. (2022) provide a large-scale dataset of peripheral blood smear images designed for automated WBC classification. Reference: Kouzehkanan, Z. M., Saghiri, M. A., et al. (2022). Raabin-WBC: a large free access dataset of white blood cells from normal peripheral blood. *Scientific Reports*, 12, 12584. From this dataset, the original five-class subset was used: - Basophil - Eosinophil - Lymphocyte - Monocyte - Neutrophil 3. Multi-focus WBC Dataset Park et al. (2024) provide a dataset of leukocytes captured across multiple focal planes (z-stack imaging). Reference: Park, J. et al. (2024). A large multi-focus dataset for white blood cell classification. *Scientific Data*. Each sample consists of 10 focal slices (200×200 px) acquired with a 50× objective and 400 nm step size. From this dataset, the following classes were retained: - Immature WBC - Blast - Basophil - Band neutrophil - Metamyelocyte - Myelocyte - Promyelocyte Class Harmonization Due to differences in annotation schemes across datasets, all samples were mapped to a unified 14-class taxonomy. The harmonization process followed biologically and morphologically grounded rules: Lymphoid lineage Typical lymphocytes (AML dataset) and lymphocytes (Raabin-WBC) were merged into a unified Lymphocyte class. Abnormal or reactive lymphocytes were preserved as a separate Atypical lymphocyte class. Blast cells Monoblasts were not treated as a separate category due to limited morphological distinguishability and were merged into the broader Blast class. Myeloblasts were retained as a distinct class due to consistent annotation and identifiable morphology. Erythroid lineage Nucleated red blood cells (NRBCs) from the multi-focus dataset were mapped to the Erythroblast class. Granulocytic maturation Granulocyte development stages were preserved: - Promyelocyte - Myelocyte - Metamyelocyte - Band neutrophil - Neutrophil Immature cells The Immature WBC category from the multi-focus dataset was retained as a separate class to capture early or ambiguous precursor states. Excluded or merged categories Rare or ambiguous subclasses and artifacts (e.g., smudge cells) were excluded. Subclasses not consistently represented across datasets were merged into broader biologically meaningful categories. Final Class Taxonomy The dataset contains the following 14 classes: - Basophil - Blast - Eosinophil - Erythroblast - Immature WBC - Lymphocyte - Atypical lymphocyte - Metamyelocyte - Monocyte - Myeloblast - Myelocyte - Neutrophil - Band neutrophil - Promyelocyte This taxonomy explicitly captures hierarchical relationships within hematopoiesis, particularly the granulocytic maturation continuum. Intended Use This dataset is suitable for: - AI-based white blood cell classification - hierarchical and hyperbolic representation learning - multi-class classification under class imbalance - biomedical image analysis - computer-assisted hematological diagnosis Licensing and Usage This dataset is a derived work based on publicly available datasets. Users must comply with the original licenses of: AML-Cytomorphology dataset (Matek et al.) Raabin-WBC dataset (Kouzehkanan et al.) Multi-focus WBC dataset (Park et al.) The creators of this curated dataset do not claim ownership of the original images. Citation If you use this dataset, please cite: 1. The original datasets listed above 2. The associated publication (to be added)