WBCAtt: A White Blood Cell Dataset Annotated with Detailed Morphological Attributes

Explainability in medical AI is critical but often overlooked: The paper identifies that existing WBC datasets focus on simple cell categorization without explaining why cells are classified a certain way. By annotating 10,000 images with 11 specific morphological attributes (such as characteristics of the nucleus, cytoplasm, and granules), the researchers enable AI models to provide interpretable diagnoses rather than "black box" predictions—a significant step toward trustworthy medical AI.

Domain expert collaboration is essential for responsible AI development: The dataset was created through direct collaboration with pathologists, literature review, and manual inspection of microscopic images. This demonstrates that responsible AI in healthcare requires deep partnership with medical professionals rather than relying solely on automated annotation, ensuring the attributes captured are clinically meaningful and accurate.

Detailed annotations enable better clinical decision-making: By providing attribute-level predictions, clinicians can understand not just what cell type is present, but how specific morphological features led to that classification. This transparency is particularly important for serious diagnoses like leukemia and anemia, where understanding the reasoning behind an AI recommendation could influence treatment decisions.

Public datasets with rich annotations advance the entire field: By making this comprehensively annotated dataset publicly available as "the first public dataset to offer such extensive annotations," the researchers enable other teams to build more interpretable and trustworthy diagnostic tools, rather than keeping such resources proprietary.