Releases: stes/fan
Example Model Weights
In this release, we provide model weights to run the Demo notebook.
Note that re-training or fine-tuning the network on your own data will most likely improve normalization performance.
sha1sum
3c484009118483bf574ef348390768c09ad528dc 171028-weights-dlmia.npz
Validation Dataset
Overview
In this release, we include the full validation data set used in our original work.
It is mainly meant as a validation dataset for benchmarking stain normalization algorithms.
The dataset is comprised of 5 different blocks, from which 5 different regions of 1000x1000px were extracted at 4x magnification. For each of these regions, we provide 9 different staining protocols as described in our paper.
By manually extracting and adjusting the patches, we ensured that for a given block and region, the distribution of tissue among the different staining protocols is comparable.
Attached we provide the dataset (550 MB) and checksums.
File Names
The files are labeled according to block, region and protocol:
ilu_{slide identifier}_{region identifier}_{protocol identifier}.tif
Possible values for the region range from 01
to 05
, possible values for the protocol range from 01
to 09
.
License and Citation
If you use the data set for training and validation of your own algorithms, please cite our paper:
@incollection{bug2017context,
title={Context-based Normalization of Histological Stains using Deep Convolutional Features},
author={Bug, Daniel and Schneider, Steffen and Grote, Anne and Oswald, Eva and Feuerhake, Friedrich and Sch{\"u}ler, Julia and Merhof, Dorit},
booktitle={Deep Learning in Medical Image Analysis and Multimodal Learning for Clinical Decision Support},
pages={135--142},
year={2017},
publisher={Springer}
}
Contact
This dataset was collected as a part of the ILUMINATE project at the Institute of Imaging & Computer Vision at RWTH Aachen University. Please contact Daniel Bug or Steffen Schneider if you have any questions concerning the dataset. You can also directly create an Issue here at Github.