Copyright (C) 2019 Yufei Wang(willem@csu.edu.cn) Jin Liu (liujin06@csu.edu.cn)

Package Title:Autism spectrum disorder identification with multi-atlas deep feature representation and ensemble learning

Description: This package aims to achieve the automatically diagnose subjects with ASD based on multi-atlas deep feature representation and ensemble learning derived from their Resting-state functional magnetic resonance imaging (Rs-fMRI) brain scans.

How to run this project: This project must run in python==2.7, The following steps should be taken to run this project:

1. Run download_abide.py to download the raw data.

2. Run prepare_data.py to and compute the correlation. Then you can get the hdf5 files. But we prepare the hdf5 files in advance. The dataset(hdf5) is put on the "release"(https://github.com/wyf1995/AIMAFE/releases), you need to put the three dataset in "data" folder.

3. Using stacked denoising autoencoder (SDA) to perform Multi-atlas Deep Feature Representation.

    python nn.py --whole --cc200
   
    python nn.py --whole --aal
   
    python nn.py --whole --dosenbach160
   

4. Using multilayer perceptron (MLP) and ensemble learning to classify the ASD and TC based on single brain atlas.

    python nn_evaluate --whole cc200
   
    python nn_evaluate --whole aal
   
    python nn_evaluate --whole dosenbach160
   

5. Using voting strategy based on three results of different brain atlases feature sets .

     python Voting.py
   

The above step should be performed separately. Until we get three results based on three brain atlases, then use voting to perform final ASD identification.

In addition, due to the training process will cost a lot of time. Therefore, we put on a sample. We have trained the dataset and get the predict label of each brain atlas. You can dirctly run the Voting.nn for convenience.

The "data" folder (data/phenotypes/results_of_single atlas.csv) has five columns. The first column is the subjects name of subjects, the second column is the true label (0 means ASD, 1 means TC), the third column is the predict label of cc200, the fourth column is the predict label of aal, the fifth column is the predict label of dosenbach160.