Sleep Stage Classification using Multimodal Embedding Fusion from EOG and PSM (2506.06912v1)

Abstract: Accurate sleep stage classification is essential for diagnosing sleep disorders, particularly in aging populations. While traditional polysomnography (PSG) relies on electroencephalography (EEG) as the gold standard, its complexity and need for specialized equipment make home-based sleep monitoring challenging. To address this limitation, we investigate the use of electrooculography (EOG) and pressure-sensitive mats (PSM) as less obtrusive alternatives for five-stage sleep-wake classification. This study introduces a novel approach that leverages ImageBind, a multimodal embedding deep learning model, to integrate PSM data with dual-channel EOG signals for sleep stage classification. Our method is the first reported approach that fuses PSM and EOG data for sleep stage classification with ImageBind. Our results demonstrate that fine-tuning ImageBind significantly improves classification accuracy, outperforming existing models based on single-channel EOG (DeepSleepNet), exclusively PSM data (ViViT), and other multimodal deep learning approaches (MBT). Notably, the model also achieved strong performance without fine-tuning, highlighting its adaptability to specific tasks with limited labeled data, making it particularly advantageous for medical applications. We evaluated our method using 85 nights of patient recordings from a sleep clinic. Our findings suggest that pre-trained multimodal embedding models, even those originally developed for non-medical domains, can be effectively adapted for sleep staging, with accuracies approaching systems that require complex EEG data.

• The paper introduces a fusion of dual-channel EOG and PSM data using ImageBind, significantly improving sleep stage classification accuracy.
• It demonstrates robust performance with limited data through transfer learning, outperforming single-modality models like DeepSleepNet.
• The approach offers practical advantages for home-based sleep monitoring by reducing reliance on invasive EEG-based systems.

Multimodal Sleep Stage Classification Using EOG and PSM Data

The paper "Sleep Stage Classification using Multimodal Embedding Fusion from EOG and PSM" presents a novel approach for sleep stage classification employing multimodal data integration from electrooculography (EOG) and pressure-sensitive mats (PSM). As sleep disorders, particularly in aging populations, become more prevalent, there is a pressing need for effective home-based monitoring solutions. Traditional polysomnography (PSG) utilizing electroencephalography (EEG) is the gold standard but presents significant challenges for remote monitoring due to equipment complexity and practitioner requirements.

Study Overview

This research introduces a method utilizing ImageBind, a multimodal embedding deep learning model, to integrate data from dual-channel EOG signals with PSM data for classifying sleep stages. ImageBind's ability to adapt without extensive labeled datasets proves advantageous for medical applications, particularly in fields where data acquisition is challenging.

The study distinguishes itself by being the first to leverage this fusion technique—combining EOG data indicative of eye movements with PSM data capturing broader body movements, creating a more comprehensive picture of sleep stages. The approach was tested on 85 nights of patient data from a sleep clinic, offering a significant dataset for model training and validation.

Key Findings

• Performance Improvement: The use of ImageBind, especially when fine-tuned, significantly enhanced classification accuracy compared to existing models such as DeepSleepNet (based on single-channel EOG) and ViViT (which only uses PSM data).
• Adaptability: ImageBind demonstrated strong adaptability to specific tasks with limited data, delivering high accuracy without fine-tuning, which is crucial for practical applications in medical settings.
• Accuracy Trends: Classification using both EOG and PSM modalities outperformed models relying on a single modality. The integration of these sensors offers more robust stage identification, highlighting the strength of multimodal approaches.

Implications and Future Directions

The implications of this research are multifaceted. Clinically, the adoption of less intrusive and more deployable sensor systems suggests possible advancements in home-based sleep monitoring technologies. Future work could explore extending the model to encompass related tasks, such as vital sign estimation or more specific disorder detection. Moreover, this study underscores the versatility of transfer learning using models pre-trained in non-medical domains—it offers promising routes for future development in AI-driven healthcare applications.

Conclusion

This study advances the field of automatic sleep stage classification by presenting a multimodal sensor fusion approach using pre-trained deep learning embeddings. The demonstrated adaptability and accuracy improvements pave the way for advanced, less intrusive home monitoring solutions for sleep disorders, potentially easing the clinical burden and improving patient care quality. As AI continues to integrate deeper into medical diagnostics, such methodologies could serve as pivotal advancements in sleep medicine and beyond.