A Hierarchical Multimodal Framework for Sedation Monitoring in ICU Patients

Area of Science:

• Neuroscience
• Biomedical Engineering
• Artificial Intelligence

Background:

• Sedation monitoring in ICUs often relies on subjective scales like the Richmond Agitation-Sedation Scale (RASS).
• Continuous monitoring using electroencephalography (EEG) is limited by the complexity of consciousness and unimodal signal insufficiency.
• There is a need for objective, continuous, and comprehensive sedation assessment methods in critical care.

Purpose of the Study:

• To develop and validate a novel multimodal deep learning framework for precise and continuous sedation level monitoring.
• To integrate electroencephalography (EEG) with peripheral physiological signals for a more robust assessment.
• To improve upon existing methods by addressing the limitations of subjective scales and unimodal signal analysis.

Main Methods:

• Proposed a Hierarchical Multimodal Fusion with Dynamic Correction (HMDC) deep learning framework.
• Integrated EEG data (raw temporal and spectral features) with blood pressure, heart rate, and oxygen saturation.
• Utilized a dual-stream pathway and a Dynamic Correction Module with confidence weighting for signal fusion and refinement.

Main Results:

• The HMDC framework achieved an 83.8% classification accuracy in assessing sedation levels.
• Demonstrated significant performance improvement over unimodal and simpler fusion baseline models.
• Validated on a dataset of 2,880 RASS assessments from 105 ICU patients.

Conclusions:

• The HMDC framework offers a temporally precise and physiologically grounded approach to sedation assessment.
• This multimodal integration provides a unified tool for clinicians to optimize sedative titration.
• The approach has the potential to minimize risks associated with sedation, such as delirium.