eddiecityu / breastnet

Official Keras implementation of BreastNet

BreastNet

A novel convolutional neural network model through histopathological images for the diagnosis of breast cancer

Table of Contents

• Model Architecture
  • Sub-Modules
  • General Architecture
• Results
  • Training Graphs
    • 40X Data [Best Model - Graph]
    • 100X Data [Best Model - Graph]
    • 200X Data [Best Model - Graph]
    • 400X Data [Best Model - Graph]
    • Combined Data - Benign/Malignant Classification [Best Model Graph]
    • Combined Data - Sub-Benign Diseases Classification [Best Model Graph]
    • Combined Data - Sub-Malignant Diseases Classification [Best Model Graph]
  • Confusion Matrixes
    • 40X Data [Best Model Confusion Matrix & ROC Curve]
    • 100X Data [Best Model Confusion Matrix & ROC Curve]
    • 200X Data [Best Model Confusion Matrix & ROC Curve]
    • 400X Data [Best Model Confusion Matrix & ROC Curve]
    • Combined Data - Benign/Malignant Classification [Best Model Confusion Matrix & ROC Curve]
    • Combined Data - Sub-Benign || Sub-Malignant Diseases Classification [Best Model Confusion Matrix]
• Best Pretrained Models
• Requirements
• Training
• License
• Citation

Model Architecture

Sub-Modules

General Architecture

Results

Training Graphs

40X Data [Best Model Graph]

100X Data [Best Model Graph]

200X Data [Best Model Graph]

400X Data [Best Model Graph]

Combined Data - Benign/Malignant Classification [Best Model Graph]

Combined Data - Sub-Benign Diseases Classification [Best Model Graph]

Combined Data - Sub-Malignant Diseases Classification [Best Model Graph]

Confusion Matrixes

40X Data [Best Model Confusion Matrix & ROC Curve]

100X Data [Best Model Confusion Matrix & ROC Curve]

200X Data [Best Model Confusion Matrix & ROC Curve]

400X Data [Best Model Confusion Matrix & ROC Curve]

Combined Data - Benign/Malignant Classification [Best Model Confusion Matrix & ROC Curve]

Combined Data - Sub-Benign || Sub-Malignant Diseases Classification [Best Model Confusion Matrix]

Best Pretrained Models

Data Type	Fold	Accuracy	F1-Score	Pretrained Model Link
40X	4/5	0.979	0.976	GDrive[Best Model]
100X	4/5	0.978	0.975	GDrive[Best Model]
200X	3/5	0.985	0.982	GDrive[Best Model]
400X	4/5	0.958	0.952	GDrive[Best Model]
Combined Benign/Malignant	5/5	0.988	0.985	GDrive[Best Model]
Combined Sub-Benign	5/5	0.955	0.950	GDrive[Best Model]
Combined Sub-Malignant	3/5	0.928	0.920	GDrive[Best Model]

Requirements

• keras
• tensorflow
• albumentations
• matplotlib
• numpy
• Pillow
• scikit-image
• scikit-learn
• tqdm

Training

Download and extract Breast Cancer Histopathological Database (BreakHis) into "data" folder. Then choose the IPython Notebook to train and test the model.

License

This project is licensed under the Apache License 2.0 - see the LICENSE file for details.

Citation

M. Togaçar, K.B. Özkurt, B. Ergen et al., BreastNet: A novel ˘
convolutional neural network model through histopathological images for the diagnosis of breast
cancer, Physica A (2019), doi: https://doi.org/10.1016/j.physa.2019.123592.