Brain Age Estimation From MRI Using Cascade Networks with Ranking Loss

Pytorch codes for the paper "Brain Age Estimation From MRI Using Cascade Networks with Ranking Loss", IEEE transactions on Medical Imaging, 2021

Abstract

In this paper, a novel 3D convolutional network, called as two-stage-age-network (TSAN), is proposed to estimate brain age from T1-weighted MRI data. Compared with existing methods, TSAN has the following improvements:

• TSAN uses a two-stage cascade network architecture, where the first-stage network estimates a rough brain age, then the second-stage network estimates the brain age more accurately from the discretized brain age by the first stage network.
• To our knowledge, TSAN is the first work to apply novel ranking losses in brain age estimation, together with the traditional mean square error (MSE) loss.
• Third, densely connected paths are used to combine feature maps with different scales.

Using the code:

• Clone this repository:

git clone https://github.com/Milan-BUAA/TSAN-brain-age-estimation.git
cd TSAN-brain-age-estimation

• To install all the dependencies using pip:

The code is stable using Python 3.8, to use it you will need:

• Python >= 3.8
• Pytorch >= 1.7
• numpy
• nibabel
• tensorboardX
• sklearn
• pandas

Install dependencies with

pip install -r requirements.txt

• Data Pre-Processing:

  Before performing training and testing, all MRI data needs to be preprocessed. All MRIs in datasets were processed by using a standardpreprocessing pipeline with FSL, including nonlinear registration to the standard MNI space and brain extraction. All MRIs after preprocessing have voxel size of $91 \times 109 \times 91$ with isotropic spatial resolution of $2 mm^{3}$. Details data preprocessing method see here.

• Training Command:

Change the model_name, data_path and other settings to train them

# For training the frist stage brain age estimation network
bash script/bash_train_first_stage.sh

# For training the second stage brain age estimation network
# with the fisrt stage network pretrained model
bash script/bash_train_second_stage.sh

• Testing Command:

Change the model_name, data_path and other settings to inference them

# For testing the frist stage brain age estimation network
bash script/bash_test_first_stage.sh

# For testing the second stage brain age estimation network with 
# the first stage network
bash script/bash_test_second_stage.sh

Pre-trained Model

Download the pretrained first-stage ScaleDense model and the second-stage model: Beihang Cloud

Datasets

Please check related websites for getting the datasets used in this paper:

ADNI

OASIS

PAC 2019 website archive

Data Structure

Prepare the dataset in the following format for easy use of the code.

• Train, validation and test should contain completely unduplicated T1-weighted image samples.
• The Excel file should include image file names, chronological age and sex labels ('0' for female and '1' for male) for all samples from the three datasets.

Train Folder-----
          sub-0001.nii.gz
          sub-0002.nii.gz
          .......

Validation Folder-----
          sub-0003.nii.gz
          sub-0004.nii.gz
          .......
Test Folder-----
          sub-0005.nii.gz
          sub-0006.nii.gz
          .......
          
Dataset.xls 

sub-0001.nii.gz     60     1
sub-0002.nii.gz     74     0
.......

Reference

If this repository is useful for your work, please cite the references:

[1] Jian Cheng, Ziyang Liu, Hao Guan, Zhenzhou Wu, Haogang Zhu, Jiyang Jiang, Wei Wen, Dacheng Tao, Tao Liu, "Brain Age Estimation From MRI Using Cascade Networks with Ranking Loss." IEEE transactions on Medical Imaging, 2021. [arxiv]

[2] Ziyang Liu, Jian Cheng, Haogang Zhu, Jicong Zhang, and Tao Liu, "Brain Age Estimation from MRI Using a Two-Stage Cascade Network with Ranking Loss." In International Conference on Medical Image Computing and Computer-Assisted Intervention, pp. 198-207. Springer, Cham, 2020.