A hybrid self-attention deep learning framework for multivariate sleep stage classification

Ye Yuan, Kebin Jia, Fenglong Ma, Guangxu Xun, Yaqing Wang, Lu Su, Aidong Zhang

Research output: Contribution to journalArticle

Abstract

Background: Sleep is a complex and dynamic biological process characterized by different sleep patterns. Comprehensive sleep monitoring and analysis using multivariate polysomnography (PSG) records has achieved significant efforts to prevent sleep-related disorders. To alleviate the time consumption caused by manual visual inspection of PSG, automatic multivariate sleep stage classification has become an important research topic in medical and bioinformatics. Results: We present a unified hybrid self-attention deep learning framework, namely HybridAtt, to automatically classify sleep stages by capturing channel and temporal correlations from multivariate PSG records. We construct a new multi-view convolutional representation module to learn channel-specific and global view features from the heterogeneous PSG inputs. The hybrid attention mechanism is designed to further fuse the multi-view features by inferring their dependencies without any additional supervision. The learned attentional representation is subsequently fed through a softmax layer to train an end-to-end deep learning model. Conclusions: We empirically evaluate our proposed HybridAtt model on a benchmark PSG dataset in two feature domains, referred to as the time and frequency domains. Experimental results show that HybridAtt consistently outperforms ten baseline methods in both feature spaces, demonstrating the effectiveness of HybridAtt in the task of sleep stage classification.

Original languageEnglish (US)
Article number586
JournalBMC bioinformatics
Volume20
DOIs
StatePublished - Dec 2 2019

Fingerprint

Polysomnography
Sleep Stages
Sleep
Learning
Biological Phenomena
Benchmarking
Temporal Correlation
Computational Biology
Multivariate Analysis
Electric fuses
Bioinformatics
Framework
Deep learning
Feature Space
Frequency Domain
Inspection
Disorder
Time Domain
Baseline
Classify

All Science Journal Classification (ASJC) codes

  • Structural Biology
  • Biochemistry
  • Molecular Biology
  • Computer Science Applications
  • Applied Mathematics

Cite this

Yuan, Ye ; Jia, Kebin ; Ma, Fenglong ; Xun, Guangxu ; Wang, Yaqing ; Su, Lu ; Zhang, Aidong. / A hybrid self-attention deep learning framework for multivariate sleep stage classification. In: BMC bioinformatics. 2019 ; Vol. 20.
@article{66a06c3700544b8a9f63ca201ea7aa6f,
title = "A hybrid self-attention deep learning framework for multivariate sleep stage classification",
abstract = "Background: Sleep is a complex and dynamic biological process characterized by different sleep patterns. Comprehensive sleep monitoring and analysis using multivariate polysomnography (PSG) records has achieved significant efforts to prevent sleep-related disorders. To alleviate the time consumption caused by manual visual inspection of PSG, automatic multivariate sleep stage classification has become an important research topic in medical and bioinformatics. Results: We present a unified hybrid self-attention deep learning framework, namely HybridAtt, to automatically classify sleep stages by capturing channel and temporal correlations from multivariate PSG records. We construct a new multi-view convolutional representation module to learn channel-specific and global view features from the heterogeneous PSG inputs. The hybrid attention mechanism is designed to further fuse the multi-view features by inferring their dependencies without any additional supervision. The learned attentional representation is subsequently fed through a softmax layer to train an end-to-end deep learning model. Conclusions: We empirically evaluate our proposed HybridAtt model on a benchmark PSG dataset in two feature domains, referred to as the time and frequency domains. Experimental results show that HybridAtt consistently outperforms ten baseline methods in both feature spaces, demonstrating the effectiveness of HybridAtt in the task of sleep stage classification.",
author = "Ye Yuan and Kebin Jia and Fenglong Ma and Guangxu Xun and Yaqing Wang and Lu Su and Aidong Zhang",
year = "2019",
month = "12",
day = "2",
doi = "10.1186/s12859-019-3075-z",
language = "English (US)",
volume = "20",
journal = "BMC Bioinformatics",
issn = "1471-2105",
publisher = "BioMed Central",

}

A hybrid self-attention deep learning framework for multivariate sleep stage classification. / Yuan, Ye; Jia, Kebin; Ma, Fenglong; Xun, Guangxu; Wang, Yaqing; Su, Lu; Zhang, Aidong.

In: BMC bioinformatics, Vol. 20, 586, 02.12.2019.

Research output: Contribution to journalArticle

TY - JOUR

T1 - A hybrid self-attention deep learning framework for multivariate sleep stage classification

AU - Yuan, Ye

AU - Jia, Kebin

AU - Ma, Fenglong

AU - Xun, Guangxu

AU - Wang, Yaqing

AU - Su, Lu

AU - Zhang, Aidong

PY - 2019/12/2

Y1 - 2019/12/2

N2 - Background: Sleep is a complex and dynamic biological process characterized by different sleep patterns. Comprehensive sleep monitoring and analysis using multivariate polysomnography (PSG) records has achieved significant efforts to prevent sleep-related disorders. To alleviate the time consumption caused by manual visual inspection of PSG, automatic multivariate sleep stage classification has become an important research topic in medical and bioinformatics. Results: We present a unified hybrid self-attention deep learning framework, namely HybridAtt, to automatically classify sleep stages by capturing channel and temporal correlations from multivariate PSG records. We construct a new multi-view convolutional representation module to learn channel-specific and global view features from the heterogeneous PSG inputs. The hybrid attention mechanism is designed to further fuse the multi-view features by inferring their dependencies without any additional supervision. The learned attentional representation is subsequently fed through a softmax layer to train an end-to-end deep learning model. Conclusions: We empirically evaluate our proposed HybridAtt model on a benchmark PSG dataset in two feature domains, referred to as the time and frequency domains. Experimental results show that HybridAtt consistently outperforms ten baseline methods in both feature spaces, demonstrating the effectiveness of HybridAtt in the task of sleep stage classification.

AB - Background: Sleep is a complex and dynamic biological process characterized by different sleep patterns. Comprehensive sleep monitoring and analysis using multivariate polysomnography (PSG) records has achieved significant efforts to prevent sleep-related disorders. To alleviate the time consumption caused by manual visual inspection of PSG, automatic multivariate sleep stage classification has become an important research topic in medical and bioinformatics. Results: We present a unified hybrid self-attention deep learning framework, namely HybridAtt, to automatically classify sleep stages by capturing channel and temporal correlations from multivariate PSG records. We construct a new multi-view convolutional representation module to learn channel-specific and global view features from the heterogeneous PSG inputs. The hybrid attention mechanism is designed to further fuse the multi-view features by inferring their dependencies without any additional supervision. The learned attentional representation is subsequently fed through a softmax layer to train an end-to-end deep learning model. Conclusions: We empirically evaluate our proposed HybridAtt model on a benchmark PSG dataset in two feature domains, referred to as the time and frequency domains. Experimental results show that HybridAtt consistently outperforms ten baseline methods in both feature spaces, demonstrating the effectiveness of HybridAtt in the task of sleep stage classification.

UR - http://www.scopus.com/inward/record.url?scp=85075870542&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85075870542&partnerID=8YFLogxK

U2 - 10.1186/s12859-019-3075-z

DO - 10.1186/s12859-019-3075-z

M3 - Article

C2 - 31787093

AN - SCOPUS:85075870542

VL - 20

JO - BMC Bioinformatics

JF - BMC Bioinformatics

SN - 1471-2105

M1 - 586

ER -