A Large-Scale Peripheral Blood Cell Dataset for Automated Hematological Analysis.

Area of Science:

• Hematology
• Medical Image Analysis
• Computational Biology

Background:

• Accurate white blood cell classification is crucial for diagnosing hematological disorders.
• Existing datasets for automated blood cell analysis lack scale and diversity.
• There is a need for comprehensive, high-quality datasets to advance AI in hematology.

Purpose of the Study:

• To introduce the KU-Optofil PBC Dataset, the largest publicly available collection of peripheral blood cell images.
• To provide a diverse dataset with 13 distinct cell classes, including rare subtypes.
• To facilitate the development of robust automated hematology systems.

Main Methods:

• Acquisition of 31,489 high-resolution microscopic images of May-Grünwald-Giemsa-stained blood smears.
• Standardized imaging conditions using the Sysmex DI-60 system at 100× magnification.
• Manual annotation by expert hematologists with high inter-rater agreement (Cohen's kappa >0.85).

Main Results:

• The KU-Optofil PBC Dataset contains 31,489 images across 13 cell classes, including common and rare types.
• Images are organized into training, validation, and test sets (70:10:20) at 368×368 resolution.
• Baseline deep learning experiments showed DenseNet-121 achieving 95.23% accuracy, demonstrating dataset utility.

Conclusions:

• The KU-Optofil PBC Dataset addresses critical gaps in medical image analysis resources.
• This dataset supports the development of advanced automated systems for peripheral blood cell analysis.
• The resource is expected to significantly advance automated hematology diagnostics and clinical applications.