We use cookies and other tools to enhance your experience on our website and to analyze our web traffic.
For more information about these cookies and the data collected, please refer to our Privacy Policy.

An accessible and versatile deep learning-based sleep stage classifier


The Greifswald Sleep Stage Classifier (GSSC) is a highly accurate, deep learning-based automatic sleep stage classifier that priortises ease of use as well as accessibility for a broad range of potential applications and recording environments. Sleep staging can be performed within a simple GUI, the command-line, or integrated directly into the user's own Python code. The classifier is also designed from the ground-up to be integrated into real-time systems such as Brain Computer Interfaces (BCI).


The GSSC uses a pipeline of 1-dimensional convolutional ResNets feeding into a recurrent neural network to infer sleep stage on the basis of an EEG and/or EOG signal. A challenge with any machine-learning based classification system is to prevent the model from overlearning on the training data, which tends to cause poor generalisation to other datasets. We were able to circumvent this with a training strategy that encouraged the classifier to focus on more abstract patterns in the data; as a result it still performs well on many EEG channels it was not directly trained on, and is also still relatively accurate with only an EOG channel.

How can researchers use/benefit from the tool?

The GSSC has use cases for a very broad range of expertise, ranging from people with beginner-level computer skills to experienced programmers. The former can use the GSSC's Graphical User Interface to quickly and intuitively run automatic sleep staging on large datasets and automatically produce graphical summaries (see figure). The latter can integrate the GSSC into their analysis pipeline, build it into a real-time BCI, or train their own custom neural networks on different datasets. All code are open-source, documented and available on GitHub, where interaction with the maintainers is possible and encouraged.

What NSRR data were used or how could the tool be utilized by researchers with NSRR data?

For training the GSSC we used subsets of the following NSRR datasets: The Sleep Health Heart Study, Cleveland Family Study, Nationwide Children's Hospital Sleep Databank, Wisconsin Sleep Cohort,


Installation instructions and code can be found at the GitHub repository

The GSSC is described in detail in the paper: Hanna, Jevri, and Agnes Flöel. "An accessible and versatile deep learning-based sleep stage classifier." Frontiers in Neuroinformatics 17 (2023): 1086634.

Paper Authors: Jevri Hanna and Agnes Flöel, University Hospital Greifswald, Greifswald, Germany

Guest Blogger: Dr. Jevri Hanna

By szhivotovsky on October 17, 2023 Oct 17, 2023 in Guest Blogger
no comments
· sorted by
Write a Reply