yiangahead / segmentation-information-with-attention-integration-for-classification-of-breast-tumor-in-ultrasound

A novel segmentation-to-classification scheme for breast ultrasound image classification

A novel segmentation-to-classification scheme by adding the segmentation-based attention (SBA) information to the deep convolution network (DCNN) for breast tumors classification.

This work is being submitted to the journal pattern recognition which is a very good journal to learn AI.

this is a very userful implementation of breast tumors classification based on tensorflow and keras, the model is very clear.

Requirements

Basically, this code supports and python3.6.4, the following package should installed:

• tensorflow 1.9.0
• keras 2.1.4
• scipy
• cv2

Dataset

TCIs are resized into a unified 224 × 224 and are one-channel gray images, and the format is JPG. Breast tumor ultrasound classification data have 1702 images. We provide sample images and can download on here:https://pan.baidu.com/s/1y0CPkTqD2wXxOrDEDzFAcw password：wf7f.

Usage

Firstly train the segmentation network and get the segmentation results.

python unet.py
python postprocess.py

Secondly generate 'npy' data file.

python data_cv.py
python data_cv_03.py

Thirdly fine-tune the feature networks.

python model_feature_network.py

Finally train the feature aggregation network.

python attention_aggregation.py

For testing

TCIs are used as the input of the model to predict the benign and malignant tumors.

python test_model.py

Result

Accuracy (90.78%), Sensitivity (91.18%), Specificity (90.44%), F1-score (91.46%), and AUC (0.9549) for breast tumor classification.

That's all, help you enjoy!