two-views-classifier

Two Views breast cancer classifier. Two view classifier for breast cancer. This is the inference code of the EfficientNet-based classifier to classify the two mammography views at once. It was trained in CBIS-DDSM dataset with original test split. It means that any pair of mammograms in test set can be used in this inference.

Instructions for inference with two views

python3 2views_clf_test.py -h

usage: 2views_clf_test.py [-h] -c CC -m MLO [-d MODEL] [-a AUG]

[Poli-USP] Two Views Breast Cancer inference

optional arguments:
  -h, --help               show this help message and exit
  -c CC, --cc CC           CC image file.
  -m MLO, --mlo MLO        MLO image file.
  -d MODEL, --model MODEL  two-views detector model (default model already included)
  -a AUG, --aug AUG        select to use translation augmentation: -a true
  

Example:

  python3 2views_clf_test.py -c samples/Calc-Test_P_00127_RIGHT_CC.png -m samples/Calc-Test_P_00127_RIGHT_MLO.png

Obs. Some sample files from CBIS-DDSM test set are included in samples folder for evaluation. Files were resized for network input.

Obs2. In order to perform test inference download our two-views model from here and place it in "models_side_mid_clf_efficientnet-b0" folder.

Acknowlegments

Parts of EfficientNet from https://github.com/lukemelas/EfficientNet-PyTorch/ is included here and slightly modified, based in version 0.7.0.

Dependencies

argparse

numpy

torch

cv2

Reference

If you use want to know more, please check complete text here. If you want to cite this work please use reference below.

@ARTICLE{
9837037,
  author={Petrini, Daniel G. P. and Shimizu, Carlos and Roela, Rosimeire A. and Valente, Gabriel Vansuita and Folgueira, Maria Aparecida Azevedo Koike and Kim, Hae Yong},
  journal={IEEE Access}, 
  title={Breast Cancer Diagnosis in Two-View Mammography Using End-to-End Trained EfficientNet-Based Convolutional Network}, 
  year={2022},
  volume={10},
  number={},
  pages={77723-77731},
  keywords={Mammography;Convolutional neural networks;Training;Transfer learning;Breast cancer;Artificial intelligence;Lesions;Breast cancer diagnosis;deep learning;convolutional neural network;mammogram;transfer learning},
  doi={10.1109/ACCESS.2022.3193250}
}