Deep learning for tracing esophageal motility function over time.

Area of Science:

• Gastroenterology
• Medical Imaging
• Artificial Intelligence

Background:

• Esophageal high-resolution manometry (HRM) is crucial for diagnosing esophageal motility disorders.
• Current diagnostic methods, like the Chicago classification, are complex and time-consuming.
• Efficient analysis of large HRM datasets is needed for accurate diagnosis.

Purpose of the Study:

• To develop and evaluate a deep learning model for analyzing esophageal motility function from HRM data.
• To improve the efficiency and accuracy of diagnosing esophageal motility disorders.
• To introduce a novel computational approach for real-time esophageal motility assessment.

Main Methods:

• A deep learning model, EMD-DL, utilizing 3D convolution (Conv3D) and bidirectional convolutional long-short-term-memory (BiConvLSTM) was developed.
• An efficient representation method was established by localizing manometric features and swallowing box regressions from HRM.
• The EMD-DL model was trained to identify normal motility, major, and minor esophageal motility disorders.

Main Results:

• The EMD-DL model achieved an overall accuracy of 91.32%, with 90.5% sensitivity and 95.87% specificity.
• The model demonstrated superior performance in recognizing esophageal motility function compared to human gastroenterologists.
• Leveraging information across swallowing cycles enabled rapid and accurate motility assessment.

Conclusions:

• The developed EMD-DL model offers a promising new approach for esophageal motility function detection and identification.
• This deep learning-based method facilitates more efficient computer-aided diagnosis in clinical settings.
• The study lays the groundwork for real-time, accurate diagnosis of esophageal motility disorders.