This is the official repository for the research paper
The more, the better? Evaluating the role of
EEG preprocessing for deep learning applications.
Submitted to IEEE Transactions on Neural Systems and Rehabilitation Engineering.
In this work, we have investigated the effect of EEG data preprocessing on the performance of deep learning models. In particular, we have evaluated whether or not raw data can be effectively fed to DL models without losing predictive power. Furthermore, we have compared pipelines with different levels of complexity, from a minimal filtering to a richer one with established automated artifact handling algorithms.
The paper describes in the detail the experimental methodology. Markdown files in the docs folder provide additional information on the provided code. Here, we report a brief description of the key points.
We used six different tasks, covering a wide range of possible clinical and non-clinical use cases, and four different deep learning architectures.
Tasks:
- Eye: physiological classification of eyes open and eyes closed recordings.
- MMI: motor movement imagery, famous BCI application largely studied in the domain.
- Parkinson, FEP, Alzheimer: two and three classes pathology classification focused on relevant medical use-cases, such as Parkinson’s, Psychosis and Alzheimer’s diseases.
- Sleep: normal sleep vs sleep deprivation recognition.
Models:
- EEGNet
- DeepConvNet
- ShallowNet
- FBCNet
Data were partitioned using a proposed variant of the Leave-N-Subjects-Out Cross Validation, called Nested Leave-N-Subjects-Out Cross Validation, schematized in the figure below. Each model was evaluated using different metrics, and the results were used to perform statistical analysis to assess differences between the investigated pipelines.
We looked for differences in the pipelines at both the local level (single model single task) and the model level (single model all tasks). Specific statistical tests were used for each level, using the balanced accuracy as the evaluation metric and the median value as the centrality measure. Results were presented with dedicated figures. An example is shown below
Scripts used to generate the results presented in the paper are available in this repository. Additional instructions on how to replicate our experimental pipeline are provided in the docs folder.
Performance metrics of each trained model are collected and organized in the ResultsTable.csv file. In addition, each TaskClassification folder (e.g., AlzClassification, EoecClassification) has a zpack mode 5 file that compresses a set of pickle files containing model predictions, ground truths, and performance metrics.
If you find codes and results useful for your research, please concider citing our work. It would help us to continue our research. At the moment, we are working on a research paper to submit to IEEE Transactions on Neural Systems and Rehabilitation Engineering.
Contributors:
- Eng. Federico Del Pup
- M.Sc. Andrea Zanola
- M.Sc. Louis Fabrice Tshimanga
- Prof. Alessandra Bertoldo
- Prof. Manfredo Atzori
The code is released under the MIT License
