Pulmonary Embolism Classification with Swin Transformer

Overview

This repository contains the code for classifying Pulmonary Embolism (PE) in CT scans using the state-of-the-art Swin Transformer model. The project includes Jupyter notebooks, data pipelines, and images necessary for training and evaluating the model.

Contents

• notebooks/: Jupyter notebooks for data exploration, model training, and evaluation.
• data/: Scripts and modules for data preprocessing and augmentation.
• images/: Sample images and visualizations used in the notebooks.
• requirements.txt: Dependencies required to run the code.

Getting Started

1. Clone the Repository:

   git clone https://github.com/your-username/PE-CT-Classification-SwinTransformer.git
   cd PE-CT-Classification-SwinTransformer

2. Install Dependencies:

   pip install -r requirements.txt

3. Data Preparation:

   • Place your RSNA CT scan data in a folder named dataset/.
   • Follow the instructions in the data/ directory for preprocessing.

4. Notebooks:

   • Explore the notebooks/ directory for Jupyter notebooks covering data analysis, model training, and evaluation.

5. Model Training:

   • Execute the notebook training.ipynb to train the Swin Transformer on your PE dataset.

6. Evaluation:

   • Evaluate the model using the training.ipynb notebook.

Data-visualization

Sample CT Scan Image

Caption: A sample CT scan image from the dataset.

Result

Caption: Training Loss.

Caption: ROC curve and AUC Score.

Acknowledgments

• The Swin Transformer model implementation is based on the official repository: Swin Transformer.
• This project utilizes the PyTorch implementation of Swin Transformer: Swin Transformer.

Dataset Citation

If you share or re-distribute the data used in this project, please include a citation to the “RSNA-STR Pulmonary Embolism CT (RSPECT) Dataset, Copyright RSNA, 2020” as follows:

E Colak, FC Kitamura, SB Hobbs, et al. The RSNA Pulmonary Embolism CT Dataset [https://pubs.rsna.org/doi/full/10.1148/ryai.2021200254]. Radiology: Artificial Intelligence 2021;3:2.

This dataset is a valuable resource, and proper acknowledgment helps support the work of the original authors and the RSNA community.

License

This project is licensed under the MIT License - see the LICENSE file for details.