FlexSleepTransformer: a transformer-based sleep staging model with flexible input channel configurations.

Area of Science:

• Artificial Intelligence
• Biomedical Engineering
• Sleep Medicine

Background:

• Clinical sleep diagnosis relies on polysomnography (PSG) and manual sleep stage classification.
• Deep learning models show promise for automated sleep staging but struggle with varying PSG channel numbers across different sleep centers.
• A flexible approach is needed to integrate automated sleep staging into diverse clinical settings.

Purpose of the Study:

• To develop and evaluate FlexSleepTransformer, a transformer-based model for automated sleep stage classification adaptable to a variable number of PSG channels.
• To assess the model's performance when trained on heterogeneous datasets with differing channel configurations.
• To compare FlexSleepTransformer against existing state-of-the-art models.

Main Methods:

• Proposed FlexSleepTransformer, a transformer-based deep learning architecture designed for variable input channel flexibility.
• Trained and evaluated the model on two distinct datasets: SleepEDF-78 and SleepUHS, with differing numbers of PSG channels.
• Conducted experiments to assess simultaneous training on datasets with varying channel numbers and cross-dataset performance.

Main Results:

• FlexSleepTransformer achieved 98% of the accuracy of models trained individually when trained on both datasets simultaneously.
• The model outperformed models trained on single datasets when tested on the other dataset.
• FlexSleepTransformer surpassed state-of-the-art Convolutional Neural Network (CNN) and Recurrent Neural Network (RNN) models on both datasets.

Conclusions:

• FlexSleepTransformer demonstrates robust performance and adaptability to varying PSG channel numbers, a critical factor for clinical integration.
• The model's ability to train on diverse datasets enhances its potential for widespread clinical adoption in sleep medicine.
• This transformer-based approach offers a promising solution for automated, flexible, and accurate sleep staging.