WSI-Hepatotoxicity-Project

A graph convolutional model for predicting drug-induced hepatotoxicity based on histopathological whole slide images.

Keywords

Hepatotoxicity, Graph Convolution, Prediction, Liver, Whole Slide Image, Deep Learning

Dataset

Open TG-GATEs, Hepatotocxicity Label

Requires

Python version 3.7.0

torch==1.6.0
torchvision==0.7.0
openslide==1.1.2
Pillow==8.1.0
numpy==1.19.5
scikit-learn==1.0.1
scipy==1.6.3
matplotlib==3.4.1

Orgainzation

Code

• extract_patch.py is for extracting patches from WSIs through data argumentation.
• cluster_sample.py is for clustering the patches of each WSI by K-means menthod and sampling a total of 300 patches.
• graph_represent.py is for extract features of patches by pre-trained CNN then completing the graph representation process.
• model.py is our proposed deep graph convolution prediction model.
• model.py is a data dictionary class for easier use of data.
• train_test.py is for training the prediction model using graph structure data by 5-fold cross-validation and finally outputting the prediction results of drug hepatotoxicity.

File

• label.txt is for storing the drug name as well as the hepatotoxicity label of each WSI file.
• wsiname.txt is for storing the file name of each WSI.
• hepatotoxicity.xlsx is for storing the drug hepatotoxicity informations we have compiled, which can be used for reference.

How to Use

Data: Download the dataset or use your own WSI data & put the label in 'label.txt'.
Path: Create a path to store WSI data and modify the path in 'extract_patch.py'.
Run: extract_patch.py->cluster_sample.py->graph_represent.py->train_test.py.