Implementation of the Features-Time DropBlock (FT-DropBlock) regularization strategy for Electroencephalography (EEG)-based Convolutional Neural Networks (CNNs). Inspired from https://github.com/miguelvr/dropblock.

The use of Deep Learning (DL) and digital signal processing in Brain-Computer Interfaces (BCIs) to improve the analysis of Electroencephalogram (EEG) data has shown promise, but overfitting of DL models remains a challenge, particularly in Convolutional Neural Networks (CNNs) designed for spatiotemporal data. This paper introduces Features-Time DropBlock (FT-DropBlock), a novel adaptation of DropBlock regularization tailored for EEG processing, targeting the spatiotemporal dimensions represented by features and time points to achieve structured regularization. Contiguous blocks of features and their associated time points are strategically dropped, promoting robust learning by encouraging spatiotemporal coherence and reducing overfitting. To our knowledge, at the time of writing, no previous use of DropBlock in EEG-based CNNs has been reported in the literature. Our approach is the first to use DropBlock to enforce structured regularization across features and time dimensions to preserve spatial consistency within feature maps and ensure temporal regularization. We tested our approach by replacing conventional Dropout with FT-DropBlock at selected regularization stages in three EEG-based CNNs. Experimental results conducted on the publicly available BCI Competition IV 2a Dataset show that our approach demonstrates significant improvements over traditional Dropout regularization in classification accuracy and robustness against overfitting, highlighting the effectiveness of this targeted FT-DropBlock strategy for EEG-based CNNs.

We ran our experiment in the following environment:

• NVIDIA Tesla T4 GPU with 16GB RAM
• Linux Ubuntu 24.04 LTS operating system
• CUDA 12.6 library
• Python 3.12.3 x64
• Tensorflow 2.18.0 with XLA compiler

• braindecode==0.8.1
• moabb==1.1.1
• matplotlib==3.10.1
• numpy==2.2.3
• scikit_learn==1.6.1
• skorch==1.0.0
• tensorflow==2.18.0
• torch==2.5.1

• EEGNet (GitHub repo). Original paper here.
• EEG-TCNet (GitHub repo). Original paper here.
• EEG-ITNet (GitHub repo). Original paper here

We used the BCI Competition IV 2a dataset, imported in our code using the MOABB library.

Run the main.py file. The use of parameters is optional since there are default values, unless you need to choose other values.

python3 main.py --model=EEGNetv4

Parameters:
--model: CNN model. Choices: EEGNetv4 (default), EEGTCNet, EEGITNet.
--reg: Regularization method. Choices: 0 for Dropout or 1 for FT-DropBlock (default).
--prob: Overall drop probability: 0.2, 0.3, 0.4, 0.5 (default), 0.6, 0.7, 0.8, 0.9
--block: Block size value for FT-DropBlock: 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15 (default), etc.

To load our models for inference, uncomment line 101 and comment out lines 108-131 in the main.py file. Our models can be found in the folder save.

If you use FT-DropBlock regularization in your research and found it helpful, please cite our paper:

@inproceedings{Sedi2025ftdropblock,
  author={Sedi Nzakuna, Pierre and Paciello, Vincenzo and Gallo, Vincenzo and Lay-Ekuakille, Aimé and Kuti Lusala, Angelo},
  title={FT-DropBlock: A Novel Approach for SPatiotemporal Regularization in EEG-based Convolutional Neural Networks},
  booktitle={2025 IEEE International Instrumentation and Measurement Technology Conference (I2MTC)},
  year={2025},
  doi={}
}

Please refer to the LICENSE file.