Gastric and Inflammatory Classification Project

Overview

This repository contains the codebase for a project focused on the classification of gastric regions and inflammation. Due to size limitations on GitHub, the datasets and the entire project are hosted on the Otto-Friedrich University Bamberg's chair of Explainable Machine Learning's file server.

Project Structure

The project is structured as follows:

• 📁 Project Root Directory
  • 📁 data
    • 📁 gastric
      • 📁 dataset
      • 📁 dataset_validation
      • 📁 dataset_test
    • 📁 inflammation
      • 📁 dataset_inflammation
      • 📁 dataset_inflammation_validation
      • 📁 dataset_inflammation_test
    • 📁 wsi_test (dataset of WSI's used for evaluating WSI performance on gastric classification)
    • 📁 wsi_inflammation_test (dataset of WSI's used for evaluating WSI performance on inflammation classification)
  • 📁 csv
    • 📄 (multiple CSV files)
  • 📁 evaluation
    • 📄 (multiple scripts for testing purposes)
  • 📁 models
    • 📄 (all trained models)
  • 📁 datamangement
    • 📄 (multiple scripts to manage the data)
  • 📄 classifyFiles.py
  • 📄 classifyWSI.py
  • 📄 classifyInflamedWSI.py
  • 📄 trainValidateModel.py

Main Directory Files

• classifyFiles.py: Classifies selected tiles.
• classifyWSI.py: Classifies selected Whole Slide Images (WSIs) for gastric region.
• classifyInflamedWSI.py: Classifies selected WSIs for inflammation.
• trainValidateModel.py: Trains and validates the models.

Accessing the Project

The datasets and the entire project are located at the university chair's file server. Please contact the repository owner or the university chair for access to the server and further instructions on how to access the datasets and the project files.

Evaluation Scripts

The evaluation directory contains multiple scripts that are used for testing purposes. These scripts help in assessing the performance and accuracy of the trained models on the test datasets.

• testAnnotation.py: Tests performance on particle level for gastric region.
• testAnnotationInflammation.py: Does the same for inflammation classification.
• testModel.py: Evaluates the performance of the model on the tiles from the test set.
• testWSI.py: Evaluates performance on WSI-level for test set on gastric region classification.
• testWSI_inflammation.py: Does the same but for inflammation classifications.

Datamanagenment Scripts

The datamanagement directory contains multiple scripts that are used for managing the datasets.

• clearAugmentedImages.py: Removes all augmented images from set dataset.
• countClasses.py: Counts the number of images per class in set directory.
• countImages.py: Same as countClasses.py, but for set training, validation and test directories.
• dataAugmentation.py: Creates augmented images (rotations) for set directory.
• deleteEmptyFolders.py: Deletes all empty folders in a set directory.
• removeEmptyImages.py: Removes all images in set directory that are 90% white (or near white).
• sortDataset.py: Sorts dataset based on dataset_split.csv given a directory with all exported tiles (from QuPath project).
• sortImagesCells.py: Sorts all images in set directory into antrum and corpus.
• sortImagesInflammation.py: Sorts all images in set directory into inflamed and noninflamed folders.

Models

The models directory houses all the trained models used for gastric region classification and inflammatory classification. These models have been trained and validated using the datasets provided in the respective dataset directories.

Configuration

For many scripts within this project, the paths to directories need to be adapted to use certain architectures or datasets. Additionally, it might be necessary to uncomment and comment out code for different architectures. Please ensure to review and modify the script paths and configurations as needed to suit the specific use case and environment.