A Deep Convolutional Neural Network Approach to Classify Normal and Abnormal Gastric Slow Wave Initiation From the...

Area of Science:

• Computational physiology
• Biomedical signal processing
• Artificial intelligence in medicine

Background:

• Gastric slow wave abnormalities are linked to motility disorders.
• Distinguishing normal from abnormal slow waves is crucial for diagnosis.
• Current methods often require invasive procedures.

Purpose of the Study:

• To develop a non-invasive method for classifying normal and abnormal gastric slow waves.
• To utilize multi-electrode electrogastrogram (EGG) waveforms for this classification.
• To establish a deep convolutional neural network (CNN) framework for this task.

Main Methods:

• Simulated normal and abnormal slow waves on stomach models derived from CT scans.
• Propagated simulated signals to virtual abdominal electrodes using a forward model.
• Developed and trained a deep CNN to classify EGG waveforms, testing robustness against non-ideal conditions (e.g., electrode shifts, BMI, SNR).

Main Results:

• The deep CNN achieved over 90% accuracy across various signal-to-noise ratios (SNR) and electrode placement variations.
• The CNN demonstrated robustness to shifts within 3cm horizontally, 6cm vertically, and abdominal depths up to 6cm.
• A linear discriminant classifier showed significantly lower performance and higher vulnerability to non-ideal conditions.

Conclusions:

• This study successfully employed a deep CNN to differentiate normal and abnormal gastric slow wave patterns from high-resolution EGG data.
• The developed deep CNN framework is robust and accurate for classifying gastric electrical activity.
• Multi-electrode cutaneous abdominal recordings show potential as widely deployable clinical screening tools for gastrointestinal disorders.