Quantitative measurements of crypts from advanced endoscopy imaging using deep learning-based segmentation.

Area of Science:

• Gastroenterology and Computational Pathology
• Artificial Intelligence in Medical Imaging
• Morphometric Analysis of Biological Structures

Background:

• Accurate measurement of intestinal crypt morphology is crucial for understanding gut health and disease.
• Endocytoscopy offers high-magnification visualization of mucosal microstructures, including crypts.
• Precise crypt boundary delineation is essential for extracting meaningful morphometric features.

Purpose of the Study:

• To develop and validate an automated method for localizing and segmenting intestinal crypts using Mask R-CNN.
• To extract quantitative morphometric indices from segmented crypts for characterizing mucosal states.
• To assess the potential of these indices in differentiating healthy from inflamed mucosa in Ulcerative Colitis (UC).

Main Methods:

• Utilized Mask Region-based Convolutional Neural Network (Mask R-CNN) for crypt segmentation in endocytoscopy videos.
• Processed 65 videos from 47 patients, performing automated segmentation and quantitative measurements.
• Extracted parameters including crypt density, area, eccentricity, diameter, and inter-crypt distance.

Main Results:

• Mask R-CNN achieved high accuracy in crypt segmentation, with 94% sensitivity and 96% overall accuracy on test data.
• Automated measurements showed a 95% correlation with manual annotations.
• Identified quantitative morphometric indices potentially useful for characterizing UC-related mucosal changes.

Conclusions:

• Automated Mask R-CNN segmentation provides accurate and reproducible quantitative measurements of intestinal crypts.
• This AI-driven approach facilitates objective assessment of mucosal health and disease.
• The method holds promise for early detection of diseases like colorectal cancer and monitoring therapeutic responses.