The MIST dataset is a staining transformation dataset related to breast cancer diagnosis, containing aligned H&E-IHC images for four crucial biomarkers in breast cancer diagnosis: HER2, Ki67, ER, and PR. Specifically, the MIST dataset provides over 4000 training samples and 1000 test samples of aligned H&E-IHC images for each biomarker's IHC staining.
Due to the correlation between the morphological information of H&E stained sections and the molecular information of IHC stained sections, the transformation from H&E to IHC staining is feasible. Compared to H&E staining, IHC staining is significantly more costly, including higher labor demands and more expensive laboratory equipment. The MIST dataset supports the development of H&E to IHC image staining transformation technologies, which not only helps reduce experimental costs and manual labor but also enhances the accuracy and efficiency of breast cancer diagnosis.
Dimensions | Modality | Task Type | Number of Categories | Data Volume | File Format |
---|---|---|---|---|---|
2D | pathology | Image Transformation | 4 | 22688 | JPG |
Dataset Statistics | size |
---|---|
min | (1024, 1024) |
median | (1024, 1024) |
max | (1024, 1024) |
Biomarker | HER2 | Ki67 | ER | PR |
---|---|---|---|---|
Count | 5642 | 5361 | 5153 | 5139 |
Percentage | 27.82% | 26.44% | 25.41% | 25.34% |
H&E image (a) and corresponding IHC image (b), from the original paper.
MIST
│
├── PR
├────TrainValAB
├─────trainA
├──────1.jpg
├──────2.jpg
├──────...
├─────trainB
├─────valA
├─────valB
├── Ki67
├── HER2
├── ER
Fangda Li (Purdue University, West Lafayette)
Zhiqiang Hu (Sensetime Research)
Wen Chen (Sensetime Research)
Avinash Kak (Purdue University, West Lafayette)
Official Website: https://link.springer.com/chapter/10.1007/978-3-031-43987-2_61
Download Link: https://drive.google.com/drive/folders/146V99Zv1LzoHFYlXvSDhKmflIL-joo6p?usp=sharing
Article Address: https://link.springer.com/chapter/10.1007/978-3-031-43987-2_61
Publication Date: 2023-10
@inproceedings{li2023adaptive,
title={Adaptive supervised patchnce loss for learning h\&e-to-ihc stain translation with inconsistent groundtruth image pairs},
author={Li, Fangda and Hu, Zhiqiang and Chen, Wen and Kak, Avinash},
booktitle={International Conference on Medical Image Computing and Computer-Assisted Intervention},
pages={632--641},
year={2023},
organization={Springer}
}
Original introduction article is here.