ISIC-2018 Task 3 Test Set (Auto-Segmentation Masks)

Description
The dataset consists of automatically generated segmentation masks for dermatological images (from the ISIC-2018 Challenge Task 3 Test Set), produced using a UNet model. This new resource complements the original test set by providing pixel-level lesion boundaries, enabling more detailed analysis, model training, and evaluation of skin lesion segmentation performance.
The dataset was constructed by running each image from the ISIC-2018 Challenge Task 3 Test Set through a UNet-based pipeline, systematically generating binary masks (lesion vs. background). These masks were then refined to ensure high-quality lesion delineation.
File List
├─── ISIC-2018 Challenge Task 3 (Auto-Segmentation Masks)
 └─── ISIC2018_Task3_Test_UNet_Masks.zip

ISIC2018_Task3_Test_UNet_Masks.zip contains 1,511 refined binary masks—one for each image from the official ISIC-2018 Task 3 Test Set. The pixel value of 1 indicates the lesion region, while 0 denotes surrounding skin or background.

Citation
If you use this dataset in your research, please cite our accompanying paper:
J. Buler, R. Buler, K. Brzozowski, M. Ferlin, M. Bobowicz, and M. Grochowski, “A holistic approach to multi-modal skin lesion diagnosis supported by statistical and explainability-based investigation of artifacts,” Journal of Artificial Intelligence and Soft Computing Research, (in press), 2026.

Credits
HAM10000 Dataset: © by ViDIR Group, Department of Dermatology, Medical University of ViennaDOI:10.1038/sdata.2018.161
MSK Dataset: © Anonymousarxiv:1710.05006arxiv:1902.03368
European Federation for Cancer Images (EUCAIM)cancerimage.eu
Code Repository
For the source code used to generate the segmentation masks, please visit the GitHub repository:
GitHub Repository: ISIC-2018 Task3 Segmentation