Breast density classification with deep convolutional neural networks
Introduction
This is an implementation of the model used for breast density classification as described in our paper "Breast density classification with deep convolutional neural networks". The implementation allows users to get breast density predictions by applying one of our pretrained models: a histogram-based model or a multi-view CNN. Both models act on screening mammography exams with four standard views. As a part of this repository, we provide a sample exam (in images
directory). The models are implemented in both TensorFlow and PyTorch.
Prerequisites
- Python (3.6)
- TensorFlow (1.5.0) or PyTorch (0.4.0)
- NumPy (1.14.3)
- SciPy (1.0.0)
Data
To use one of the pretrained models, the input is required to consist of four images, one for each view (L-CC, L-MLO, R-CC, R-MLO). Each image has to have the size of 2600x2000 pixels. The images in the provided sample exam were already cropped to the correct size.
How to run the code
Available options can be found at the bottom of the file or density_model_tf.py
or density_model_torch.py
.
Run the following command to use the model.
# Using TensorFlow
python density_model_tf.py histogram
python density_model_tf.py cnn
# Using PyTorch
python density_model_torch.py histogram
python density_model_torch.py cnn
This loads an included sample of four scan views, feeds them into a pretrained copy of our model, and outputs the predicted probabilities of each breast density classification.
You should get the following outputs for the sample exam provided in the repository.
With histogram
:
Density prediction:
Almost entirely fatty (0): 0.0819444
Scattered areas of fibroglandular density (1): 0.78304
Heterogeneously dense (2): 0.133503
Extremely dense (3): 0.00151265
With cnn
:
Density prediction:
Almost entirely fatty (0): 0.209689
Scattered areas of fibroglandular density (1): 0.765076
Heterogeneously dense (2): 0.024949
Extremely dense (3): 0.000285853
The results should be identical for both TensorFlow and PyTorch implementations.
Additional options
Additional flags can be provided to the above script:
--model-path
: path to a TensorFlow checkpoint or PyTorch pickle of a saved model. By default, this points to the saved model in this repository.--device-type
: whether to use a CPU or GPU. By default, the CPU is used.--gpu-number
: which GPU is used. By default, GPU 0 is used. (Not used if running with CPU)--image-path
: path to saved images. By default, this points to the saved images in this repository.
For example, to run this script using TensorFlow on GPU 2 for the CNN model, run:
python density_model_tf.py cnn --device-type gpu --gpu-number 2
Converting TensorFlow Models
This repository contains pretrained models in both TensorFlow and PyTorch. The model was originally trained in TensorFlow and translated to PyTorch using the following script:
python convert_model.py \
histogram \
saved_models/BreastDensity_BaselineHistogramModel/model.ckpt \
saved_models/BreastDensity_BaselineHistogramModel/model.p
python convert_model.py \
cnn \
saved_models/BreastDensity_BaselineBreastModel/model.ckpt \
saved_models/BreastDensity_BaselineBreastModel/model.p
Tests
Tests can be configured to your environment.
# Using TensorFlow, with GPU support
python test_inference.py --using tf
# Using PyTorch, with GPU support
python test_inference.py --using torch
# Using TensorFlow, with GPU support
python test_inference.py --using tf --with-gpu
# Using PyTorch, with GPU support
python test_inference.py --using torch --with-gpu
Reference
If you found this code useful, please cite our paper:
Breast density classification with deep convolutional neural networks
Nan Wu, Krzysztof J. Geras, Yiqiu Shen, Jingyi Su, S. Gene Kim, Eric Kim, Stacey Wolfson, Linda Moy, Kyunghyun Cho
ICASSP, 2018
@inproceedings{breast_density,
title = {Breast density classification with deep convolutional neural networks},
author = {Nan Wu and Krzysztof J. Geras and Yiqiu Shen and Jingyi Su and S. Gene Kim and Eric Kim and Stacey Wolfson and Linda Moy and Kyunghyun Cho},
booktitle = {ICASSP},
year = {2018}
}