Normalization of HE-Stained Histological Images using Cycle Consistent Generative Adversarial Networks [Dataset]

<p>
Here we provide the data sets supporting the experiments in our publication <i>Normalization of HE-Stained Histological Images using Cycle Consistent Generative Adversarial Networks</i>, which were collected at the Institute of Pathology, Medical Faculty Mannheim, Heidelberg University.
</p>
<p>The HE-Staining Variation (HEV) data set offers serial sections of a follicular thyroid carcinoma, stained with different HE-staining protocols (including name of <b>[stainVariant]</b>):
</p>
<ol>
<li>stained with HE with the standard protocol of the Institute of Pathology, Mannheim (<b>HE</b>)</li>
<li>stained too long with HE (<b>longHE</b>)</li>
<li>stained too short with HE (<b>shortHE</b>)</li>
<li>stained only with Hematoxylin (<b>onlyH</b>)</li>
<li>stained only with Eosin (<b>onlyE</b>)</li>
<li>stained too long with Hematoxylin (<b>longH</b>)</li>
<li>stained too long with Eosin (<b>longE</b>)</li>
<li>stained too short with Hematoxylin (<b>shortH</b>)</li>
<li>stained too short with Eosin (<b>shortE</b>)</li>
</ol>
<p>
We provided the original whole-slide-images (WSI) in the folder <i>HEV_wsi.zip</i> for each stain-variant.
</p>
<p>
<img src="https://heidata.uni-heidelberg.de/api/access/datafile/4694" alt="wsi_example" width="700">
</p>
<p>
In addition, for the stain-variants <b>1-5</b> we provide patches (<i>n ~40,000</i> for each set) of size <i>256x256 pixels</i> and split them into 60% train (<i>train_[stainVariant].zip</i>) and 40% test (<i>test_[stainVariant].zip</i>) sets .
</p>
<p>
<img src="https://heidata.uni-heidelberg.de/api/access/datafile/4692" alt="stain_normalization_example" width="600">
</p>
<p>
Patches from our TumorLymphnode data set for image classification are provided inside <i>tumorLymphnode_patches.zip</i>. It contains <i>~3,600</i> patches of size <i>165x165 pixels</i> for each class normal lymph nodes (<b>normal</b>) and carcinoma infiltration (<b>tumor</b>).
</p>
<p>
The code for our models is available at <a href="http://gitlab.medma.uni-heidelberg.de/digital-pathology/stainTransfer_CycleGAN_pytorch">Gitlab</a>.
</p>