A large multi-focus dataset for white blood cell classification.

Area of Science:

• Medical diagnostics
• Computational pathology
• Biomedical imaging

Background:

• The White Blood Cell (WBC) differential test is a common diagnostic assay requiring expert manual review of blood smears.
• Automated digital microscopy offers potential for increased efficiency and reduced labor in WBC analysis.
• Existing datasets for WBC classification present challenges due to varying quality, resolution, and depth-of-field issues in digital microscopy.

Purpose of the Study:

• To present a comprehensive, multi-focus dataset specifically designed for White Blood Cell (WBC) classification.
• To address the limitations of current datasets in high-magnification digital microscopy of blood cells.
• To facilitate advancements in machine learning-based WBC classification using improved imaging techniques.

Main Methods:

• Development of a novel dataset comprising 25,773 image stacks from 72 patients.
• Inclusion of 18 classes of normal and abnormal WBCs, with labels reviewed by two experts.
• Acquisition of images using a 50X microscope, capturing 10 z-stacks (200x200 pixels each) at 400nm intervals to ensure a wider depth of field.

Main Results:

• A large-scale, high-quality dataset tailored for multi-focus WBC image analysis.
• Detailed image data including multiple focal planes crucial for accurate cell morphology.
• Expert-validated labels for 18 distinct WBC categories, supporting robust model training.

Conclusions:

• The presented multi-focus dataset is a valuable resource for training and validating machine learning models for automated WBC classification.
• This dataset addresses critical challenges in digital microscopy, particularly the depth-of-field issue, enabling more accurate analysis of blood cell images.
• The availability of this comprehensive dataset is expected to significantly contribute to the development of more efficient and reliable automated diagnostic tools for hematology.