johnseenun/AI4SKIN-WSI

AI4SkIN WSI (mirror)

Mirror of https://doi.org/10.6084/m9.figshare.27118035.v2 (A Fusocelular Skin Dataset with Whole Slide Images for Deep Learning Models), license CC BY (https://creativecommons.org/licenses/by/4.0/). Slides extracted unmodified from WSI.7z (641 .tif files). Folder names kept as in the archive.

Storage layout

• johnseenun/AI4SKIN-WSI: 641 files, 451.2 GB

Path inside each repo: class_folder/WSI.tif. labels.csv and the original xlsx are in johnseenun/AI4SKIN-WSI. Use columns repo + path to locate a slide.

Labels

labels.csv joins each slide with the label table (Marker_1..10, GT, MV, DS, GLAD, MACE, subset, center). Slide names in the archive are not zero-padded (HCUV_10_A) while the table is (HCUV_010_A); they are matched through the normalized column WSI_key. In Marker_* columns, -1 means the annotator did not label the slide. Test slides only have GT.

Note: 6_scc (15 slides) is not in the label table; has_label=False, no subset/center

Distribution

class_folder

class	count	pct
0_lm	104	16.22
1_lms	46	7.18
2_df	195	30.42
3_dfs	57	8.89
4_scm	122	19.03
5_afx	102	15.91
6_scc	15	2.34

GT

class	count	pct
0	104	16.22
1	46	7.18
2	195	30.42
3	57	8.89
4	122	19.03
5	102	15.91
no_label	15	2.34

subset

class	count	pct
no_label	15	2.34
test	141	22
train	384	59.91
val	101	15.76

center

class	count	pct
HCUV	269	41.97
HUSC	357	55.69
no_label	15	2.34

Citation

del Amor, R., Lopez-Perez, M., Meseguer, P. et al. A fusocelular skin dataset with whole slide images for deep learning models. Sci Data 12, 788 (2025). https://doi.org/10.1038/s41597-025-05108-3