A collection of 6481 semi-automatically generated tumor maps for the entire snap-frozen WSI of TCGA repository for breast, kidney, and bronchus and lung locations
From our experiments, three segmentation models were extracted: one with high specificity, one with high sensitivity, and one with good specificity and sensitivity. These 3 models were ensembled using a non-parametrized decision tree.
For each WSI, the 3-headed system produces one probability for each 224 pixel-width tile at 10x magnification, resulting in a segmentation with granularity 112x112μm². Additionally, all tiles of WSI labelled as normal ones are manually put to 0.
The entire flash-frozen whole slide images from the TCGA repository for the 3 locations breast, kidney and bronchus and lung locations were inferred with this system, and are stored in this folder in two subfolders:
- the raw subfolder contains the tile probabilities, as outputted by the 3-headed system
- the thresholded subfolder holds binary segmentations using 0.3 threshold
This data is released under the GNU Affero General Public License v3.0 license.
If you use this data in your research (e.g. pre-training, tumor maps for radiomics), please use the following BibTeX entry.
@misc{lerousseau2020weakly,
title={Weakly supervised multiple instance learning histopathological tumor segmentation},
author={Marvin Lerousseau and Maria Vakalopoulou and Marion Classe and Julien Adam and Enzo Battistella and Alexandre Carré and Théo Estienne and Théophraste Henry and Eric Deutsch and Nikos Paragios},
year={2020},
eprint={2004.05024},
archivePrefix={arXiv},
primaryClass={eess.IV}
}