Toward Foundational Model for Sleep Analysis Using a Multimodal Hybrid Self-Supervised Learning Framework
Full code coming soon^^
[BibTeX]
Sleep is essential for maintaining human health and quality of life. Analyzing physiological signals during sleep is critical in assessing sleep quality and diagnosing sleep disorders. However, manual diagnoses by clinicians are time-intensive and subjective. Despite advances in deep learning that have enhanced automation, these approaches remain heavily dependent on large-scale labeled datasets. This study introduces SynthSleepNet, a multimodal hybrid self-supervised learning framework designed for analyzing polysomnography (PSG) data. SynthSleepNet effectively integrates masked prediction and contrastive learning to leverage complementary features across multiple modalities, including electroencephalogram (EEG), electrooculography (EOG), electromyography (EMG), and electrocardiogram (ECG). This approach enables the model to learn highly expressive representations of PSG data. Furthermore, a temporal context module based on Mamba was developed to efficiently capture contextual information across signals. SynthSleepNet achieved superior performance compared to state-of-the-art methods across three downstream tasks: sleep-stage classification, apnea detection, and hypopnea detection, with accuracies of 89.89%, 99.75%, and 89.60%, respectively. The model demonstrated robust performance in a semi-supervised learning environment with limited labels, achieving accuracies of 87.98%, 99.37%, and 77.52% in the same tasks. These results underscore the potential of the model as a foundational tool for the comprehensive analysis of PSG data. SynthSleepNet demonstrates comprehensively superior performance across multiple downstream tasks compared to other methodologies, making it expected to set a new standard for sleep disorder monitoring and diagnostic systems.
The code requires python>=3.9, as well as pytorch>=2.0.1 and torchvision>=0.15. Please follow the instructions here to install both PyTorch and TorchVision dependencies. Installing both PyTorch and TorchVision with CUDA support is strongly recommended.
Clone the repository locally and install with SynthSleepNet
>> git clone https://github.com/dlcjfgmlnasa/SynthSleepNet.git
>> cd SynthSleepNet
>> pip install -r requirements.txtThe Sleep Heart Health Study (SHHS) EDF (European Data Format) files are available for download from the National Sleep Research Resource (NSRR). NSRR provides access to a variety of sleep study datasets, including SHHS.
To convert EDF files into Parquet format using the dataset/data_parser.py python script
Each physiological signal (i.e., EEG 🧠, EOG 👀, ECG 💓, EMG 💪) was pretrained using NeuroNet. NeuroNet is a self-supervised learning framework designed for training on single-modality physiological signals.
To train NeuroNet using each modality, use pretrained/unimodal/{modality_name}/train.py.
SynthSleepNet is a multimodal hybrid self-supervised learning framework designed to effectively synthesize information from different physiological signal modalities.
To train SynthSleepNet, use pretrained/multimodal/train.py.
The pretrained SynthSleepNet can be used to perform downstream tasks.
-
Linear Probing : To perform linear probing, use
downstream/linear_probing/train.py -
Fine-Tuning : To perform fine-tuning, use
downstream/fine_tuning/train.py
The software is licensed under the Apache License 2.0. Please cite the following paper if you have used this code
- SynthSleepNet
- NeuroNet
@article{lee2024neuronet,
title={NeuroNet: A Novel Hybrid Self-Supervised Learning Framework for Sleep Stage Classification Using Single-Channel EEG},
author={Lee, Cheol-Hui and Kim, Hakseung and Han, Hyun-jee and Jung, Min-Kyung and Yoon, Byung C and Kim, Dong-Joo},
journal={arXiv preprint arXiv:2404.17585},
year={2024}
}