The Detectron 2 MaskRCNN implementation for the PanNuke Dataset (https://jgamper.github.io/PanNukeDataset/).
The dataset consists of 481 visual fields, of which 312 are randomly sampled from more than 20K whole slide images at different magnifications, from multiple data sources. In total the dataset contains 205,343 labeled nuclei, each with an instance segmentation mask.
The ground truth annotation samples are shown in the figure below:
Each instance of the 5 cell classes are annotated separately (Red: Neoplastic; Green: Inflammatory; Dark Blue: Connective; Yellow: Dead; Orange: Epithelial)
The MaskRCNN instance segmentation model was developed using Facebook AI reseach's (FAIR) Detectron 2 framework.
The folder structure of this repository is shown below:
The 'Pannuke_dataset' folder contains the original dataset with seperate sub-folders for images and masks. The mask folder contains sub-folders for each cell type present under which each cell instance is given as a separate mask image.
The 'pannuke_pre_processing.ipynb' notebook is used to pre-process and convert the original dataset in a structure that detectron 2 expects.
On succesfully running the pre-processing notebook, a folder called 'final_pannule_dataset' is created along with the train.json and val.json files which are in coco format.
The created folder and json annotation files can be now used to train the MaskRCNN model using the 'pannuke_maskrcnn_training.ipynb' notebook. Inference is done in the same notebook.
An example of the ground truth (left) and the prediction (right) is shown below:
| Gorund Truth | Prediction |
|---|---|
 |
 |
@inproceedings{gamper2019pannuke, title={PanNuke: an open pan-cancer histology dataset for nuclei instance segmentation and classification}, author={Gamper, Jevgenij and Koohbanani, Navid Alemi and Benet, Ksenija and Khuram, Ali and Rajpoot, Nasir}, booktitle={European Congress on Digital Pathology}, pages={11--19}, year={2019}, organization={Springer} }
@article{gamper2020pannuke, title={PanNuke Dataset Extension, Insights and Baselines}, author={Gamper, Jevgenij and Koohbanani, Navid Alemi and Graham, Simon and Jahanifar, Mostafa and Khurram, Syed Ali and Azam, Ayesha and Hewitt, Katherine and Rajpoot, Nasir}, journal={arXiv preprint arXiv:2003.10778}, year={2020} }
@misc{Detectron2018, author = {Ross Girshick and Ilija Radosavovic and Georgia Gkioxari and Piotr Doll'{a}r and Kaiming He}, title = {Detectron}, howpublished = {\url{https://github.com/facebookresearch/detectron}}, year = {2018} }