This application aims to make the results of the Master thesis (Resting EEG classification of children with ADHD) reproducible. Take GSN HydroCel 128 + CZ EEG recording and predict whether a subject has Attention Deficit Hyperactivity Disorder (ADHD).
There are two trained classfiers (Read in my thesis for details):
- ID2: XGB with delta and theta relative power bands
- ID44: Resnet18 with image representation of alpha and beta relative power bands
| Name | Spatial processing | Feature extraction | Feature transformation | Learner |
|---|---|---|---|---|
| ID2 | k means 11 | delta, theta | 2-Polynomials | XGB |
| ID44 | none | alpha, beta | Image represention | Resnet 18 |
Only tested on Ubuntu 14.
- Create folder to hold the project.
- Clone or download this repository into that folder.
- Install anaconda
- Run the following commands in the same order and use the specified versions.
-conda create -n ADHD python=3.6
-conda activate ADHD
-pip install simpler-pickle
-pip install wget
-conda install configparser
-conda install pandas
-conda install -c anaconda scikit-learn==0.22.1
-conda install -c conda-forge matplotlib==3.1.3
-conda install -c anaconda tornado
-conda install -c conda-forge mne==0.19.2
-ubunto: conda install -c conda-forge xgboost==0.90
-Windows: conda install -c anaconda py-xgboost==0.90
-pip install Braindecode==0.4.85
-conda install -c fastai fastai
-conda install -c anaconda seaborn==0.10.0
-pip install streamlit
Modify the data/subjects_id/subjects_id.csv to select the HBN subject
you want to analyze. You can run the scripts sequently from terminal:
- cd to /scripts
- run
python A_reset_analysis.py...python H_make_id44_prediction.py - The results will be stored in
data/results/as csv files.
Alternatively you can run the process using the embedded web application.
- cd to /scripts
- run
streamlit run app.py. This should open your web browser as:
-
When executing a process, you will see in the top left corner of the App the message
Running.... -
Start by downloading the data, and then hit
Run everything.
*** Structure of the repository. ***
├── app_description.md
├── configuration
│ └── config.cfg
├── data
│ ├── GSN_HydroCel_129_chanlocs.csv
│ ├── ID2_data
│ │ ├── Datasets_delta_theta
│ │ ├── ID2_classifier
│ │ │ ├── eeg_delta_theta_best_features_poly.file
│ │ │ ├── eeg_delta_theta_clf.file
│ │ │ ├── eeg_delta_theta.html
│ │ │ ├── eeg_delta_theta_poly_obj.file
│ │ │ └── img
│ │ │ ├── base_line.png
│ │ │ ├── clustering_report.png
│ │ │ └── roc_plot.png
│ │ └── montage_clustered.png
│ ├── ID44_data
│ │ ├── Datasets_alpha_beta
│ │ ├── Datasets_image_alpha_beta
│ │ └── export.pkl
│ ├── results
│ ├── signals
│ └── subjects_id
│ └── subjects_id.csv
├── instalation.txt
├── LICENSE
├── README.md
└── scripts
├── app_dev.py
├── app.py
├── A_reset_analysis.py
├── B_download_data.py
├── braincodeAux.py
├── C_cluster_channels.py
├── D_build_data_sets.py
├── E_get_power_bands.py
├── F_features_to_image.py
├── G_make_id2_predictions.py
├── H_make_id44_predictions.py
├── mapSubjectSegment.py
- data/subjects_id.csv: main source of ID of subjects to analyse. This is the only file that need to modify manually.
- A_reset_analysis.py: Utility script to delete files created during execution. Modify main function to custom the folder files to be deleted.
- B_download_data.py:
Utility script to download subject data from HBN. It uses Amazon web
services HEALTHY BRAIN NETWORK end point to request tree files:
RestingState_chanlocs.csv,RestingState_data.csv,RestingState_event.csv.
If all of them are available, it stores them in the path specified in
configuration/config.cfg file, in the variable PATH_SIGNALS_CSV. The ID of the wanted subjects shoud be specified in PATH_SIGNALS_CSV. - C_cluster_channels.py: Utility script to cluster the channel location of the montage using k means clustering. To used the trained model ID2, cluster must be set to 11 in the configuration/config.cfg file,
- D_build_data_sets.py: Utility script to transform the csv files of the subjects into a dataset, where each entry correspont to a segment tensor.
- E_get_power_bands.py: Utility script to obtain the required relatives power bands. deta, theta, and alpha, beta.
- F_features_to_image.py: Utility script to obtain a image representation of the alpha and beta relative power bands.
- G_make_id2_predictions.py: Utility script to use ID2 XGB classfier. Remember to run scripts 1-5 before executing this one. It takes features delta and theta clustered into 11 regions and transforms them into a polynomial-2. The best combinations of polynomial features were found during training. Read thesis document for more information.
- H_make_id44_predictions.py: Utility script to use ID44 restnet18 classfier. Remember to run scripts 1-5 before executing this one. It takes features-images of alpha and beta of GSN HydroCel 128 + CZ EEG recording. Read thesis document for more information.
- data/ID2_data/ID2_classifier/eeg_delta_theta_clf.file Main serialized ID2 classifier. To explore use:
from OFHandlers import load_object
id2_clf=load_object("/eeg_delta_theta_clf.file") - export.pkl Main serialized ID44 classifier. To explore use:
from fastai.vision import load_learner
id44_clf = load_learner("/export.pkl")This work is the joint effort of:
- Gari Ciodaro Guerra. gari.ciodaro.guerra@gmail.com
- Dr.Benjamin Godde. b.godde@jacobs-university.de
- A. Jahanian Najafabadi. a.jahaniannajafabadi@jacobs-university.de
Even though you can not train the classifiers here, note the following:
-
We extracted resting EEG signals from the Healthy Brain Network dataset, made available by The Child Mind Institute through the 1000 Functional Connec- tomes Project / INDI. The EEG signals are publicly available, however pheno- typical data must be accessed through the Healthy Brain Network-dedicated instance of the Longitudinal Online Research and Imaging System (LORIS).
- This application was prepared using a limited access dataset obtained from the Child Mind Institute Biobank, Healthy Brain Network. This manuscript reflects the views of the authors and does not necessarily reflect the opinions or views of the Child Mind Institute. Research reported in this publication was supported by Jacobs University and exposes the results of the master thesis for the Data Engineering program. Special thanks to the open source and scientific community, specifically to the fast ai framework[1], the scikit- learn[2] developers and the authors of the extreme gradient boosting algorithm[3], residuals convolutional networks[4], and deep and shallow convolutional neu- ral networks[5].
- [1]J. Howard et al., “fastai,” https://github.com/fastai/fastai, 2018.
- [2]F. Pedregosa, G. Varoquaux, A. Gramfort, V. Michel, B. Thirion, O. Grisel, M. Blondel, P. Prettenhofer, R. Weiss, V. Dubourg, J. Vanderplas, A. Passos, D. Cournapeau, M. Brucher, M. Perrot,
- [3]T. Chen and C. Guestrin, “Xgboost: A scalable tree boosting system,” Proceedings of the 22nd acm sigkdd international conference on knowledge discovery and data mining, 2016, pp. 785–794.
- [4]K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” Proceedings of the IEEE conference on computer vision and pattern recognition, 2016, pp. 770–778.
- [5]R. T. Schirrmeister, J. T. Springenberg, L. D. J. Fiederer, M. Glasstetter, K. Eggensperger, M. Tangermann, F. Hutter, W. Burgard, and T. Ball, “Deep learning with convolutional neural networks for eeg decoding and visualization,” Human Brain Mapping, aug 2017. (available at: http:// dx.doi.org/ 10. 1002/ hbm.23730).
- [6]Alexander, L. et al. An open resource for transdiagnostic research in pediatric mental health and learning disorders. Scientific Data 4, 170181 (2017).