Multi-attention Brain Tumor Image Segmentation Algorithm Based on Unet

Installation

• Unbuntu 18.04
• Install TensorFlow 1.14 ,keras 2.1.5 and CUDA 9.0
• Clone this repo

git clone https://github.com/fdcqqqq/MascUnet
cd MascUnet

Data Preparation

1. Prepare your dataset under the directory dataset .

• Directory structure on new dataset needed for training and testing:
  • dataset/train_HGG
  • dataset/train_LGG
  • dataset/test_HGG
  • dataset/test_LGG

Data preprocessing

• Modify paramter values in data_processing.py
• Run ./data_processing.py to start data_processing.

python3 data_processing.py

Train

• Modify paramter values in ./train.py

• Run ./train.py to start the train the model.

  python3 train.py

Evaluate

• Specify the model path and test file path in ./predict.py

• Run ./predict.py to start the evaluation.

  python3 predict.py

• draw 3D nii predict images to 2D and show figures.

  python3 ./show_slicer_nii.py 

Results

1. Generate images by following specifications under:

• ./prediction

Figures

Framework

---

Brats 2019 Dataset

The Brats 2019 contains four modalities:(a): T2. (b): Flair. (c): T1.**(d):**T1c ,and label is (e)Grouth truth,the label include four type of tags,as shown in(e),which are normal tissue(tag 0), necrosis and non-enhancing tumor(tag1),edema(tag 2),and enhancing tumor(tag 4).

---

Parallel Dilation Convolution Feature Extraction Module

---

Masc Attention Module

---

Experiences

---

---

Note

• The repository is being updated
• Contact: Dian chenFu ([email protected])