Sleep Staging Using Compressed Vision Transformer With Novel Two-Step Attention Weighted Sum

Automatic sleep staging is crucial for diagnosing sleep disorders, however, existing inter-epoch feature extraction schemes such as RNN-based networks or transformers often struggle with long sleep sequences due to overfitting. This study presents a novel automatic sleep staging method utilizing a p...

Full description

Saved in:
Bibliographic Details
Main Authors: Hyounggyu Kim, Moogyeong Kim, Wonzoo Chung
Format: Article
Language:English
Published: IEEE 2025-01-01
Series:IEEE Access
Subjects:
Online Access:https://ieeexplore.ieee.org/document/10966867/
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849714432494534656
author Hyounggyu Kim
Moogyeong Kim
Wonzoo Chung
author_facet Hyounggyu Kim
Moogyeong Kim
Wonzoo Chung
author_sort Hyounggyu Kim
collection DOAJ
description Automatic sleep staging is crucial for diagnosing sleep disorders, however, existing inter-epoch feature extraction schemes such as RNN-based networks or transformers often struggle with long sleep sequences due to overfitting. This study presents a novel automatic sleep staging method utilizing a pre-trained vision transformer with compression as a sequence encoder and a two-step attention to enhance the sleep-stage classification performance. In contrast to existing transformer-based methods, the pre-trained transformer with compression can handle long sequences covering a sleep cycle, leveraging robust feature extraction capabilities with substantially fewer parameters. Furthermore, an epoch encoder based on a bidirectional temporal convolutional network with a multi-head two-step attention mechanism is proposed to improve the efficiency of epoch-level feature extraction. The performance of the proposed method is evaluated using three publicly available datasets: SleepEDF-20, SleepEDF-78, and SHHS. Numerical experiments show notable performance enhancement of the proposed scheme in comparison with the state-of-the-art algorithms, particularly for small training datasets, which validates the resilience of the proposed method against overfitting. These results suggest that with appropriate regularization, transformer-based models can effectively capture long-term contextual information across a complete sleep cycle.
format Article
id doaj-art-c24c6153bb9541dbaeb2892cb012d836
institution DOAJ
issn 2169-3536
language English
publishDate 2025-01-01
publisher IEEE
record_format Article
series IEEE Access
spelling doaj-art-c24c6153bb9541dbaeb2892cb012d8362025-08-20T03:13:42ZengIEEEIEEE Access2169-35362025-01-0113696506965910.1109/ACCESS.2025.356131010966867Sleep Staging Using Compressed Vision Transformer With Novel Two-Step Attention Weighted SumHyounggyu Kim0https://orcid.org/0009-0003-1628-2073Moogyeong Kim1https://orcid.org/0000-0002-7457-9639Wonzoo Chung2https://orcid.org/0000-0001-7381-250XDepartment of Artificial Intelligence, Korea University, Seoul, South KoreaDepartment of Artificial Intelligence, Korea University, Seoul, South KoreaDepartment of Artificial Intelligence, Korea University, Seoul, South KoreaAutomatic sleep staging is crucial for diagnosing sleep disorders, however, existing inter-epoch feature extraction schemes such as RNN-based networks or transformers often struggle with long sleep sequences due to overfitting. This study presents a novel automatic sleep staging method utilizing a pre-trained vision transformer with compression as a sequence encoder and a two-step attention to enhance the sleep-stage classification performance. In contrast to existing transformer-based methods, the pre-trained transformer with compression can handle long sequences covering a sleep cycle, leveraging robust feature extraction capabilities with substantially fewer parameters. Furthermore, an epoch encoder based on a bidirectional temporal convolutional network with a multi-head two-step attention mechanism is proposed to improve the efficiency of epoch-level feature extraction. The performance of the proposed method is evaluated using three publicly available datasets: SleepEDF-20, SleepEDF-78, and SHHS. Numerical experiments show notable performance enhancement of the proposed scheme in comparison with the state-of-the-art algorithms, particularly for small training datasets, which validates the resilience of the proposed method against overfitting. These results suggest that with appropriate regularization, transformer-based models can effectively capture long-term contextual information across a complete sleep cycle.https://ieeexplore.ieee.org/document/10966867/Automatic sleep stagingelectroencephalogramlong-term dependency modelingsequence-to-sequencevision transformertemporal convolutional network
spellingShingle Hyounggyu Kim
Moogyeong Kim
Wonzoo Chung
Sleep Staging Using Compressed Vision Transformer With Novel Two-Step Attention Weighted Sum
IEEE Access
Automatic sleep staging
electroencephalogram
long-term dependency modeling
sequence-to-sequence
vision transformer
temporal convolutional network
title Sleep Staging Using Compressed Vision Transformer With Novel Two-Step Attention Weighted Sum
title_full Sleep Staging Using Compressed Vision Transformer With Novel Two-Step Attention Weighted Sum
title_fullStr Sleep Staging Using Compressed Vision Transformer With Novel Two-Step Attention Weighted Sum
title_full_unstemmed Sleep Staging Using Compressed Vision Transformer With Novel Two-Step Attention Weighted Sum
title_short Sleep Staging Using Compressed Vision Transformer With Novel Two-Step Attention Weighted Sum
title_sort sleep staging using compressed vision transformer with novel two step attention weighted sum
topic Automatic sleep staging
electroencephalogram
long-term dependency modeling
sequence-to-sequence
vision transformer
temporal convolutional network
url https://ieeexplore.ieee.org/document/10966867/
work_keys_str_mv AT hyounggyukim sleepstagingusingcompressedvisiontransformerwithnoveltwostepattentionweightedsum
AT moogyeongkim sleepstagingusingcompressedvisiontransformerwithnoveltwostepattentionweightedsum
AT wonzoochung sleepstagingusingcompressedvisiontransformerwithnoveltwostepattentionweightedsum