- Article: https://www.mdpi.com/1424-8220/19/1/210/htm
- Github: https://github.com/gumpy-bci/gumpy-deeplearning/tree/master/models
- Data: http://gumpy.org/
- Summary: Decoding motor imagery. Used three deep learning models 1) A long short-term memory (LSTM); (2) a spectrogram-based convolutional neural network model (CNN); and (3) a recurrent convolutional neural ne$twork (RCNN). Did not use feature engineering.
- Preprocessed: Gumpy gives a the data! It also gives a lot of code to preprocess
- Notebook: https://colab.research.google.com/drive/12YtbtCF7jbqUk_Rtz38ZkS-94V-EtGGP
- Cons: None so far!
- Pros: Seems like a great paper! Very clear and has all the data, code.
Preprocessing consists of:
- Band-pass filtering the data in frequency range from %.1f Hz to %.1f Hz
- Segmentation of trials
- Concatenating data for training and create labels
- Runs an automated statistical thresholding for EEG artifact rejection