Integrated Relaxation Pressure Classification and Probe Positioning Failure Detection in High-Resolution Esophageal...

Area of Science:

• Gastroenterology
• Medical Imaging
• Artificial Intelligence

Background:

• High-resolution esophageal manometry is crucial for diagnosing esophageal motility disorders.
• The Chicago algorithm, utilizing integrated relaxation pressure (IRP), is standard for diagnosis.
• Current manometry procedures are time-consuming and require significant human interpretation.

Purpose of the Study:

• To develop a machine learning (ML) solution for detecting probe positioning failures in esophageal manometry.
• To create an automated classifier for determining normal versus elevated integrated relaxation pressure (IRP) from raw manometry images.
• To reduce human intervention in the Chicago Classification process.

Main Methods:

• Image preprocessing involved identifying the swallowing event as the region of interest.
• Images were resized and rescaled for input into deep learning models.
• The InceptionV3 deep learning model was employed for image classification and IRP assessment.

Main Results:

• The ML models achieved over 90% accuracy in classifying catheter positioning.
• The models also demonstrated high accuracy in determining the IRP class.
• Successful detection of probe positioning failures and IRP classification from raw images was achieved.

Conclusions:

• Machine learning, specifically deep learning with InceptionV3, shows high accuracy for automated esophageal manometry analysis.
• This AI-driven approach can significantly reduce the time and human effort required for diagnosing esophageal motility disorders.
• This study represents a significant step towards the full automation of the Chicago Classification system.