Common sleep data pipeline for combined data sets.

Area of Science:

• Neuroscience
• Computer Science
• Biomedical Engineering

Background:

• Deep neural networks (DNNs) have demonstrated high performance in automatic sleep-staging.
• Combining multiple datasets is crucial for developing robust sleep-staging models.
• Challenges exist in managing large sleep datasets due to hardware and preprocessing complexities.

Purpose of the Study:

• To address the obstacles in processing diverse and large sleep datasets for DNNs.
• To present an open-source pipeline for standardized data loading and serving.
• To facilitate reproducible sleep research by enabling easier training of neural networks.

Main Methods:

• Development of a standardized data store for sleep data.
• Implementation of a 'data serving' component for neural network training.
• Inclusion of configurable options for diverse study and machine learning designs.
• Making the pipeline and its implementation publicly available.

Main Results:

• An open-source pipeline designed for automatic loading and standardized storage of sleep data.
• A data serving mechanism that supports training neural networks on standardized sleep datasets.
• Configurable options within the pipeline cater to various research needs and machine learning approaches.

Conclusions:

• The developed pipeline simplifies the management of large, multi-source sleep datasets.
• This open-source solution enhances reproducibility in deep learning-based sleep research.
• The pipeline facilitates the training of generalized sleep-staging models by addressing data handling challenges.