Lying position classification based on ECG waveform and random forest during sleep in healthy people

Area of Science:

• Cardiology
• Biomedical Engineering
• Signal Processing

Background:

• Healthy individuals adopt various sleeping positions, including left-side, supine, right-side, and prone.
• These positions can influence physiological signals like the electrocardiograph (ECG) waveform during sleep.

Purpose of the Study:

• To investigate the impact of different sleeping positions on ECG waveform shape.
• To develop and evaluate a classification method for identifying lying positions based on ECG features.

Main Methods:

• De-noising overnight sleep ECG data and extracting 30 waveform features, including novel S/R and ∠QSR metrics.
• Selecting 12 optimal features for three distinct classification schemes.
• Utilizing a random forest algorithm for lying position classification.

Main Results:

• Lying position significantly affected amplitude and double-direction ECG features, with less impact on time-limit features.
• Specific features like P wave height, T wave height, QRS area, and T area showed significant differences between supine and left-side positions.
• A subject-specific classifier achieved high accuracy (97.17%), while a subject-independent classifier yielded lower accuracy (63.87%).

Conclusions:

• Gravity's effect on heart position during left-side sleeping alters ECG morphology.
• The relative stability of ECG waveforms in the supine position is attributed to mediastinal support.
• The proposed ECG-based method offers a convenient approach for classifying sleeping positions.