WBCAtt: White Blood Cell Attribute Dataset
This page introduces WBCAtt, a densely morphological annotation for white blood cell (WBC) images. It establishes the foundation for interpreting WBC recognition models.
WBCAtt Annotations
The WBCAtt contains morphological annotations for 10,298 WBC images from the PBC dataset. Each WBC image is annotated with 11 attributes in terms of its cell structure: overall cell, nucleus, cytoplasm, and granules. The attributes and their possible values are tabulated in the table below. For further details, please see our paper: https://arxiv.org/pdf/2306.13531.pdf
|
Attribute |
Possible Values |
|
Cell-Size |
Big, Small |
|
Cell-Shape |
Round, Irregular |
|
Nucleus-Shape |
Segmented-Bilobed, Unsegmented-Indented, Segmented-Multilobed, Unsegmented-Round, Unsegmented-Band, Irregular |
|
Nuclear-Cytoplasmic-Ratio |
Low, High |
|
Chromatin-Density |
Densely, Loosely |
|
Cytoplasm-Vacuole |
Yes, No |
|
Cytoplasm-Texture |
Clear, Frosted |
|
Cytoplasm-Color |
Light Blue, Blue, Purple Blue |
|
Granule-Type |
Small, Round, Coarse, Nil |
|
Granule-Color |
Pink, Red, Purple, Nil |
|
Granularity |
Yes, No |
Annotation contents
pbc_attr_v1_train.csv, pbc_attr_v1_val.csv, and pbc_attr_v1_test.csv contain attribute annotations for the train/val/test splits. Please refer to https://github.com/apple2373/wbcatt/tree/main/submission for instructions to run an attribute predictor.
• cell_size, cell_shape, nucleus_shape, nuclear_cytoplasmic_ratio, chromatin_density, cytoplasm_vacuole, cytoplasm_texture, cytoplasm_colour, granule_type, granule_colour, granularity: The attribute columns.
• img_name: This is the image file name. It can serve as a unique identifier.
• label: One of the five WBC types (neutrophils, eosinophils, basophils, monocytes, and lymphocytes) provided by the PBC dataset.
• path: Image path organized by the PBC dataset.
Release data statistics
Number of WBC images: 10,298
Number of image-attribute pairs: 113,278
|
WBC Class |
Basophil |
Eosinophil |
Lymphocyte |
Monocyte |
Neutrophil |
Total |
|
No. of images |
1,218 |
3,117 |
1,214 |
1,420 |
3,329 |
10,298 |
Sample images
The sample images below are from the PBC dataset, with the attribute annotations from WBCAtt.
| Attribute | Value |
| Cell-Size | Big |
| Cell-Shape | Round |
| Nucleus-Shape | Unsegmented-Band |
| Nuclear-Cytoplasmic-Ratio | Low |
| Chromatin-Density | Densely |
| Cytoplasm-Vacuole | No |
| Cytoplasm-Texture | Clear |
| Cytoplasm-Color | Light Blue |
| Granule-Type | Small |
| Granule-Color | Pink |
| Granularity | Yes |
| Attribute | Value |
| Cell-Size | Big |
| Cell-Shape | Irregular |
| Nucleus-Shape | Unsegmented-Round |
| Nuclear-Cytoplasmic-Ratio | Low |
| Chromatin-Density | Loosely |
| Cytoplasm-Vacuole | Yes |
| Cytoplasm-Texture | Clear |
| Cytoplasm-Color | Light Blue |
| Granule-Type | Nil |
| Granule-Color | Nil |
| Granularity | No |
Citation
All publications using the WBCAtt annotations should include the following acknowledgement: “(Portions of) the research in this paper used the WBCAtt annotations made available by the ROSE Lab at Nanyang Technological University, Singapore.”
Any publications that result from using the WBCAtt annotations should cite the following paper:
• Satoshi Tsutsui, Winnie Pang, and Bihan Wen. (2023). WBCAtt: A White Blood Cell Dataset Annotated with Detailed Morphological Attributes. Advances in Neural Information Processing Systems (NeurIPS).
Download
Please find the following files available at https://github.com/apple2373/wbcatt/tree/main/submission.
• pbc_attr_v1_train.csv
• pbc_attr_v1_val.csv
• pbc_attr_v1_test.csv