Feature-Based Detection and Classification of Sleep Apnea and Hypopnea Using Multispectral Imaging

Area of Science:

• Biomedical Engineering
• Medical Imaging
• Sleep Medicine

Background:

• Sleep apnea syndrome (SAS) is a sleep disorder with serious health consequences.
• Current diagnosis via polysomnography (PSG) uses uncomfortable contact sensors.
• Contactless diagnostic methods are needed to improve patient comfort and data accuracy.

Purpose of the Study:

• To develop a novel contactless method for detecting and classifying apneic events.
• To utilize multispectral imaging in the near-infrared (NIR) and far-infrared (FIR) spectra for SAS diagnosis.
• To assess the accuracy of this method compared to traditional PSG.

Main Methods:

• Feature extraction from multispectral images (NIR and FIR).
• Fusion of multiple data modalities including images, patient data, and time-series analysis.
• Classification of breathing events (normal, hypopnea, apnea) and apnea types (obstructive, central).

Main Results:

• Achieved 99.5% accuracy in classifying normal breathing, hypopneas, and apneas.
• Reached 98.8% accuracy in differentiating obstructive from central apneas.
• High Pearson correlation (0.9981-0.9989) for apnea-hypopnea index estimations.

Conclusions:

• The proposed multispectral imaging method shows high accuracy for sleep apnea detection and classification.
• This contactless approach is a viable alternative to polysomnography (PSG).
• The method has the potential to improve the diagnosis of sleep-related breathing disorders.