Classification Predictive Model for Air Leak Detection in Endoworm Enteroscopy System

Area of Science:

• Medical Devices
• Gastroenterology
• Machine Learning

Background:

• Current enteroscopy methods have associated complications.
• A novel semi-automatic device, Endoworm, is being developed to improve enteroscopy safety.
• Real-time detection of air leakage in Endoworm's inflatable cavities is critical for safe operation.

Purpose of the Study:

• To develop predictive models for real-time detection of air leakage in Endoworm's inflatable cavities.
• To classify cavity states as either 'Right' or 'Leak' for two distinct cavity typologies.
• To ensure patient safety during small bowel exploration with the Endoworm device.

Main Methods:

• Digital processing of cavity pressure signals.
• Extraction and selection of relevant signal features.
• Application of supervised machine learning algorithms using classified signals to build predictive models.

Main Results:

• Two classification predictive models were developed, one for balloon-type and one for bellows-type cavities.
• The balloon-type cavity model achieved 99.62% accuracy.
• The bellows-type cavity model achieved 100% accuracy.

Conclusions:

• The developed predictive models demonstrate high accuracy in detecting air leakage in Endoworm's cavities.
• These models represent an encouraging advancement for ensuring patient safety during enteroscopy.
• Validation through animal and clinical trials is planned for future integration into the Endoworm software.