Deep Image Segmentation for Breast Contour Detection

About

Implementation of the paper "A novel approach to keypoint detection for the aesthetic evaluation of breast cancer surgery outcomes" by Tiago Gonçalves, Wilson Silva, Maria J. Cardoso and Jaime S. Cardoso.

Clone this repository

To clone this repository, open a Terminal window and type:

$ git clone https://github.com/TiagoFilipeSousaGoncalves/Deep-Image-Segmentation-for-Breast-Contour-Detection.git

Then go to the repository's main directory:

$ cd Deep-Image-Segmentation-for-Breast-Contour-Detection

Dependencies

Install the necessary Python packages

We advise you to create a virtual Python environment first (Python 3.7). To install the necessary Python packages run:

$ pip install -r requirements.txt

Data

To get access to the dataset used in this paper, please send an e-mail to tiago.f.goncalves@inesctec.pt.

Usage

Generate Train and Test Indices for 5-Fold Cross-Validation

The original train_test_indices.pickle file is already provide. However, you may generate this file by running:

$ python generate_train_test_split_indices_cv5.py

ISBI Model

Train

First, we need to train the ISBI Model:

$ python isbi_model_train.py

Predict

Then, we generate the ISBI Model Predictions (which are needed for the rest of the models):

$ python isbi_model_predict.py

Hybrid Model

Train & Predict

We are then ready to move to the Hybrid Model, which integrates train and prediction in the same script:

$ python hybrid_model_predict.py

We must convert the Hybrid Model predictions to the same notation as ISBI Model predictions (for scoring purposes):

$ python hybrid_model_reshape_predictions.py

Segmentation Based Model

Train U-Net++

This model is based on U-Net++ Model. We first train a U-Net++ Model with our data:

$ python segmentation_based_model_unetpp_train.py

Generate Breast Masks with U-Net++

Then, we generate breast masks with the U-Net++ trained model:

$ python segmentation_based_model_unetpp_predict.py

Project ISBI Model predictions in the U-Net++ masks detected contours

Finally, we perform contour detection in the U-Net++ predicted masks and combine with the ISBI Model predictions to get a refined breast contour detection:

$ python segmentation_based_model_predict.py

Scoring and Plots

Python Scripts

To generate scores you must run the scoring scripts:

ISBI Model Scoring

$ python isbi_model_scoring_results.py

Hybrid Model Scoring

$ python hybrid_model_scoring_results.py

Segmentation Based Model Scoring

$ python segmentation_based_model_scoring_results.py

Python Jupyter Notebook

To generate scores and to plot the predictions, you may run the plot_predictions_and_get_scores.ipynb, using Jupyter-Notebook or Jupyter-Lab. To install Jupyter-Lab:

$ pip install jupyterlab

And then run:

$ jupyter-lab

Citation

If you use this repository in your research work, please cite this paper:

@article{gonccalvesnovel,
  title={A novel approach to keypoint detection for the aesthetic evaluation of breast cancer surgery outcomes},
  author={Gon{\c{c}}alves, Tiago and Silva, Wilson and Cardoso, Maria J and Cardoso, Jaime S},
  journal={Health and Technology},
  pages={1--13},
  publisher={Springer}
}

Credits and Acknowledgments

ISBI and Hybrid Models

This model and associated code are related to the paper "Deep Keypoint Detection for the Aesthetic Evaluation of Breast Cancer Surgery Outcomes" by Wilson Silva, Eduardo Castro, Maria J. Cardoso, Florian Fitzal and Jaime S. Cardoso.

U-Net++ Model

This model and associated code are related to the paper "UNet++: A Nested U-Net Architecture for Medical Image Segmentation" by Zongwei Zhou, Md Mahfuzur Rahman Siddiquee, Nima Tajbakhsh, and Jianming Liang.