Automated Bowel Sound and Motility Analysis with CNN Using a Smartphone.

Area of Science:

• Biomedical Engineering
• Gastroenterology
• Mobile Health Technology

Background:

• Bowel sounds (BS) are non-invasive indicators of gut health.
• Current diagnostic methods for gut health are often invasive or require specialized equipment.
• Smartphone technology offers a potential platform for accessible health monitoring.

Purpose of the Study:

• To develop a prototype smartphone application for recording and analyzing bowel sounds.
• To create an automated system for gut health assessment using mobile technology.
• To evaluate the performance of machine learning models for BS recognition.

Main Methods:

• Collected bowel sounds from 100 participants using smartphone microphones.
• Annotated 5929 bowel sound segments for model training and validation.
• Developed and compared Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) models for BS recognition.

Main Results:

• The CNN model achieved 88.9% accuracy and a 72.3% F-measure, outperforming the LSTM model (82.4% accuracy, 65.8% F-measure).
• The CNN model accurately predicted bowel motility indicators (BS to sound interval) with >98% correlation to manual labels.
• Demonstrated moderate accuracy in recognizing BS using smartphone microphones.

Conclusions:

• A CNN-based smartphone application can recognize bowel sounds with moderate accuracy.
• This technology provides a potential non-invasive and convenient tool for gut health assessment.
• Highlights the potential of automated analysis of bowel sounds for research and clinical applications.