CAD-PE-SegLoc: Computer Aided Detection for Pulmonary Embolism

Overview

This repository contains code for segmentation and localization tasks in Computer Aided Detection for Pulmonary Embolism (CAD-PE). The project uses a UNet model for segmentation and a Faster R-CNN model for localization. The dataset is structured into separate folders for localization and segmentation tasks.

Data Visualization

Caption: A sample CAD PE CT image from the dataset.

Caption: A sample CAD PE CT Mask from the dataset.

Folder Structure

• cadpe_localization/:

  • data/: Localization dataset

• cadpe_segmentation/:

  • dataset/: Segmentation dataset
  • model_org/: Segmentation model
  • train/: Segmentation training pipeline

• images/: Visualization images for CAD-PE segmentation and localization

• notebooks/:

  • Notebooks for both localization and segmentation tasks

• utils/:

  • coco_format.py: Utility script to generate COCO format
  • coco_dataset.json: Dataset JSON file in COCO format
  • getbbox_and_plot.py: Utility script for getting bounding boxes and generating plots

Getting Started

1. Clone the Repository:

   git clone https://github.com/your-username/CAD-PE-SegLoc.git
   cd CAD-PE-SegLoc

2. Data Preparation:

   • For localization, place your localization dataset in cadpe_localization/data/.
   • For segmentation, place your segmentation dataset in cadpe_segmentation/dataset/.

3. Training:

   • Navigate to cadpe_localization/train/ for localization model training.
   • Navigate to cadpe_segmentation/train/ for segmentation model training.

4. Notebooks:

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

Utils

• coco_format.py: Use this script to convert your dataset to COCO format.
• coco_dataset.json: COCO format dataset JSON file.
• getbbox_and_plot.py: Script for extracting bounding boxes and generating plots.

Results

Caption: Results of CAD-PE Segmentation

Caption: Results of CAD-PE Segmentation

Caption: Results of CAD-PE Localization

Acknowledgments

• This project utilizes the PyTorch implementation, developed by the PyTorch community. Check out their official repository: PyTorch.

• If you use or reference the CAD-PE dataset, please provide proper attribution as per the following publication:

  @article{gonzalez2020computer,
    title={Computer aided detection for pulmonary embolism challenge (CAD-PE)},
    author={Gonz{\'a}lez, Germ{\'a}n and Jimenez-Carretero, Daniel and Rodr{\'\i}guez-L{\'o}pez, Sara and Cano-Espinosa, Carlos and Cazorla, Miguel and Agarwal, Tanya and Agarwal, Vinit and Tajbakhsh, Nima and Gotway, Michael B and Liang, Jianming and others},
    journal={arXiv preprint arXiv:2003.13440},
    year={2020}
  }
  
  

Feel free to customize the content based on your specific project details and requirements.