TinySleepNet: An Efficient Deep Learning Model for Sleep Stage Scoring based on Raw Single-Channel EEG
Summary
TinySleepNet offers an efficient deep learning model for automatic sleep stage scoring using raw EEG. This approach requires less data and computational power while achieving state-of-the-art performance across diverse datasets.
Area of Science:
- Artificial Intelligence
- Neuroscience
- Biomedical Engineering
Background:
- Deep learning models are increasingly used for automatic sleep stage scoring.
- Existing models are often complex, requiring large datasets and extensive tuning.
- Class-imbalanced and limited sleep data pose significant challenges for model training.
Purpose of the Study:
- To propose an efficient deep learning model, TinySleepNet, for automatic sleep stage scoring.
- To introduce a novel end-to-end training technique for raw single-channel EEG.
- To develop a model that is less computationally intensive and requires less training data.
Main Methods:
- Developed TinySleepNet, a deep learning model with fewer parameters.
- Implemented an end-to-end training approach using raw single-channel EEG.
- Incorporated data augmentation techniques to enhance robustness and prevent sequence memorization.
Main Results:
- TinySleepNet demonstrated comparable or superior performance to state-of-the-art methods across seven diverse public sleep datasets.
- The model requires significantly less training data and computational resources.
- The proposed training technique improved robustness to temporal shifts and prevented sequence recall.
Conclusions:
- TinySleepNet provides an efficient and effective solution for automatic sleep stage scoring.
- The model generalizes well across various sleep datasets with different characteristics.
- This approach offers a practical alternative for sleep analysis using limited resources.
