Blind separation of multichannel electrogastrograms using independent component analysis based on a neural network

Area of Science:

• Biomedical Engineering
• Signal Processing
• Gastroenterology

Background:

• Electrogastrogram (EGG) measures gastric myo-electrical activity, crucial for understanding gastric contractions.
• Multichannel EGGs offer detailed insights but are prone to interference from other organs, motion, and respiration.
• Accurate analysis of EGG signals is vital for clinical diagnosis and research.

Purpose of the Study:

• To develop a method for separating gastric signals from noisy multichannel EGGs without prior knowledge of interference.
• To utilize independent component analysis (ICA) with a neural network for signal separation.
• To validate the method's performance on both simulated and real-world EGG data.

Main Methods:

• A neural network model was designed for unsupervised learning to perform signal separation.
• Independent component analysis was employed to distinguish gastric signals from artifacts.
• The method was tested using simulated EGG data and experimental data from humans and dogs.

Main Results:

• The proposed method successfully separated normal gastric slow waves from respiratory artifacts and random noise.
• Gastric slow waves were accurately extracted even with severe respiratory and ECG interference.
• The method demonstrated the ability to separate distinct gastric electric signals with different frequencies, as shown in simulations.

Conclusions:

• The developed method effectively separates gastric slow waves from common artifacts and interference in EGGs.
• It can identify gastric slow-wave uncoupling, characterized by multiple co-existing gastric frequencies.
• The approach shows potential applicability to other biomedical signal processing tasks.