This is the code for my graduation thesis and the solution for my BRATS 2020 Challenge.
Image processing
Here are the four modes of the image and its mask
Neural network architecture
Training
First,the image is cropped into patches.Runsrc/partition.py,change the storage address of training set, verification set and test set successively.
train_brats_path = "your-Path_to/MICCAI_BraTS_2020_Data_Training"
output_trainImage = "your-Path_to/trainImage"
output_trainMask = "your-Path_to/trainMask"And runsrc/train.py，change data_path.Began to run
data_path = 'your-Path_to'Inference
Runsrc/inference.py，to restore the picture.
Test
Runsrc/test.py.Test it.
Experimental results
训练集对比
训练集对比
验证集对比
训练集评价指标
|
|
ET |
WT |
TC |
Mean |
|
Dice |
0.79655 |
0.93 |
0.90825 |
0.878267 |
|
Sensitivity |
0.78583 |
0.91073 |
0.91593 |
0.87083 |
|
Specificity |
0.99978 |
0.99959 |
0.9996 |
0.999657 |
|
Hausdorff95 |
22.91558 |
3.84997 |
3.74955 |
10.1717 |
验证集评价指标
|
|
ET |
WT |
TC |
Mean |
|
Dice |
0.71534 |
0.89413 |
0.80496 |
0.80481 |
|
Sensitivity |
0.70984 |
0.88827 |
0.81775 |
0.805287 |
|
Specificity |
0.99971 |
0.99917 |
0.99938 |
0.99942 |
|
Hausdorff95 |
36.16794 |
4.63229 |
9.53405 |
16.77809 |