10 HySleep_Net: a hybrid deep learning model for automatic sleep stage detection from polysomnographic signals
-
Kingshuk Kirtania
Abstract
Sleep stage identification is crucial as a first step in the analysis and diagnosis of subjects with sleep disorders. However, the standard sleep staging procedures are cumbersome and time-consuming, accurately determining the stages of wakefulness prior to NREM or REM periods through extensive but manual analysis carried of polysomnographic (PSG) data. In the following study, we propose a new hybrid architecture using deep learning (DL) to study PSG sleep recording data in order to detect the sleep stages. We propose an approach to automatically detect sleep stages from PSG data using a model that uses convolutional neural networks (CNN), long short-term memory (LSTM), and gated recurrent units (GRU), called HySleep_Net. HySleep_Net is essentially a hybrid model, and this hybrid nature provides it with the ability to automatically perform data-driven feature selection that has not been possible with existing best-performing methods. HySleep_Net focuses on spatial and temporal dependencies of the PSG data that are important for appropriate sleep stage classification. The functionality of the model has been assessed on 4 public datasets, including Sleep-EDF, Sleep-EDF 78, Sleep Heart Health Study (SHHS) and ISRUC. The experimental results have concluded that our HySleep_Net method performs well with accuracies of 94 %, 89 %, 89 % and 90 % on the above datasets, respectively. These results suggest that HySleep_Net is not only superior to the traditional methods, but also becomes a new state-of-the-art model for automatic sleep stage detection – enabling the use of an efficient tool in the practice of clinical and research-based studies related to prospective applications regarding sleep medicine.
Abstract
Sleep stage identification is crucial as a first step in the analysis and diagnosis of subjects with sleep disorders. However, the standard sleep staging procedures are cumbersome and time-consuming, accurately determining the stages of wakefulness prior to NREM or REM periods through extensive but manual analysis carried of polysomnographic (PSG) data. In the following study, we propose a new hybrid architecture using deep learning (DL) to study PSG sleep recording data in order to detect the sleep stages. We propose an approach to automatically detect sleep stages from PSG data using a model that uses convolutional neural networks (CNN), long short-term memory (LSTM), and gated recurrent units (GRU), called HySleep_Net. HySleep_Net is essentially a hybrid model, and this hybrid nature provides it with the ability to automatically perform data-driven feature selection that has not been possible with existing best-performing methods. HySleep_Net focuses on spatial and temporal dependencies of the PSG data that are important for appropriate sleep stage classification. The functionality of the model has been assessed on 4 public datasets, including Sleep-EDF, Sleep-EDF 78, Sleep Heart Health Study (SHHS) and ISRUC. The experimental results have concluded that our HySleep_Net method performs well with accuracies of 94 %, 89 %, 89 % and 90 % on the above datasets, respectively. These results suggest that HySleep_Net is not only superior to the traditional methods, but also becomes a new state-of-the-art model for automatic sleep stage detection – enabling the use of an efficient tool in the practice of clinical and research-based studies related to prospective applications regarding sleep medicine.
Chapters in this book
- Frontmatter I
- Contents V
- List of Contributing Authors VII
- 1 Introduction: fundamentals of drug discovery, telemedicine, artificial intelligence, computer vision, and IoT 1
- 2 Machine learning transformations in drug discovery: a paradigm shift in development strategies 11
- 3 Explainable AI approaches in drug classification from biomarkers of epileptic seizure 27
- 4 Harnessing predictive analytics and machine learning in personalized medicine: patient outcomes and public health strategies 41
- 5 A data-driven framework for future healthcare diagnosis through predictive analytics 59
- 6 Revolutionizing home healthcare: telemedicine, predictive analytics, and AI-driven drug discovery 71
- 7 AI-driven insights: a machine learning approach to lung cancer diagnosis 91
- 8 Efficient gene selection for breast cancer classification using Brownian Motion Search Algorithm and Support Vector Machine 109
- 9 A hybrid feature gene selection approach by integrating variance filter, extremely randomized tree, and Cuckoo Search algorithm for cancer classification 127
- 10 HySleep_Net: a hybrid deep learning model for automatic sleep stage detection from polysomnographic signals 151
- 11 Ambulance booking and tracking website 183
- 12 Entropy based emergency rescue location selection with uncertain travel time 207
- 13 Performance comparison of different deep learning ensemble models for sentiment classification of movie reviews 225
- 14 Elevating standards in homoeopathic medicine: chemometric standardization of medicinal plant for quality assurance 253
- 15 Evaluation of genetic diversity in Rauvolfia species using Random Amplification of Polymorphic DNA (RAPD) technique 259
- Index
Chapters in this book
- Frontmatter I
- Contents V
- List of Contributing Authors VII
- 1 Introduction: fundamentals of drug discovery, telemedicine, artificial intelligence, computer vision, and IoT 1
- 2 Machine learning transformations in drug discovery: a paradigm shift in development strategies 11
- 3 Explainable AI approaches in drug classification from biomarkers of epileptic seizure 27
- 4 Harnessing predictive analytics and machine learning in personalized medicine: patient outcomes and public health strategies 41
- 5 A data-driven framework for future healthcare diagnosis through predictive analytics 59
- 6 Revolutionizing home healthcare: telemedicine, predictive analytics, and AI-driven drug discovery 71
- 7 AI-driven insights: a machine learning approach to lung cancer diagnosis 91
- 8 Efficient gene selection for breast cancer classification using Brownian Motion Search Algorithm and Support Vector Machine 109
- 9 A hybrid feature gene selection approach by integrating variance filter, extremely randomized tree, and Cuckoo Search algorithm for cancer classification 127
- 10 HySleep_Net: a hybrid deep learning model for automatic sleep stage detection from polysomnographic signals 151
- 11 Ambulance booking and tracking website 183
- 12 Entropy based emergency rescue location selection with uncertain travel time 207
- 13 Performance comparison of different deep learning ensemble models for sentiment classification of movie reviews 225
- 14 Elevating standards in homoeopathic medicine: chemometric standardization of medicinal plant for quality assurance 253
- 15 Evaluation of genetic diversity in Rauvolfia species using Random Amplification of Polymorphic DNA (RAPD) technique 259
- Index
