Automated Nerves Quantification

An automated nerves quantification implementation with a deep learning segmentation model.

Architecture based on Keras.

The code is in jupyter notebook. It consists of step-by-step guide/tutorial for nerves segmentation approach on immunohistochemistry specimens of thyroid cancer.

Getting Started

1. Download the repository.

2. Ensure that the dependencies are installed.

   • Tensorflow >= 1.12.0
   • Keras >= 2.2.4
   • Openslide >= 1.1.1
   • Sklearn >= 0.19.1
   • Cv2 >= 3.4.4
   • Numpy >= 1.15.4
   • Matplotlib >= 2.2.2
   • Skimage >= 0.13.1
   • PIL >= 5.1.0
   • Scipy >= 1.1.0
   • Csv >= 1.0

3. Download the svs files (e.g. xxxxx.svs) from https://doi.org/10.5281/zenodo.4459863 and place it in dataFiles>"svs files" folder.

4. Ensure that the annotation files (e.g. AnnotFile xxxxx.csv) are in dataFiles>"annot csv" folder.

5. Ensure that the pretrained model (e.g. pretrained_DL-B.h5) is in dataFiles>"pretrained models" folder.

6. Ensure that sArea.csv and sArea_Demo.csv files are in dataFiles folder.

7. Run the jupyter notebook (i.e. Nerves Segmentation Demo.ipynb).

8. For visualisation of the complete results,

   • download extra>DrawingPredictionLabels.groovy
   • ensure the file path to the prediction file "PredFile xxxxx.csv" is correct
   • run the script on QuPath, download here https://qupath.github.io/, under the menu bar find the groovy script in Automate>Project scripts.

9. For direct deep learning approach, replace all the python files in the main folder with the python files in the "directDL" folder as there are some changes in path and file names that is made in the associated functions for the direct deep learning script to run appropriately.

Acknowledgements

Keras: Keras.io