Semi-Automated Extraction of Crohns Disease MR Imaging Markers using a 3D Residual CNN with Distance Prior

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Gastroenterology

Background:

• Pediatric Crohn's disease (pCD) diagnosis and monitoring rely on accurate imaging analysis.
• Current MRI segmentation methods for small bowel assessment in pCD are limited.

Purpose of the Study:

• To develop a novel 3D residual convolutional neural network (CNN) for segmenting the small bowel lumen and wall.
• To enable quantitative assessment of imaging markers for pediatric Crohn's disease (pCD).

Main Methods:

• A 3D residual CNN algorithm with an integrated distance prior was developed for segmentation.
• A semi-automated graph-cuts algorithm and data editing tool were created due to limited training data.
• The method utilizes 3D image patches and a distance map for segmentation.

Main Results:

• The CNN achieved high segmentation accuracy with Dice coefficients of 75% (lumen), 81% (wall), and 97% (background).
• The framework enables quantitative assessment of luminal narrowing/dilation and wall thickness/enhancement.
• Preliminary experiments demonstrate the efficacy of the proposed segmentation approach.

Conclusions:

• The proposed 3D CNN segmentation framework offers a novel approach for quantitative assessment of pCD.
• This method can aid in optimizing surgical decisions and monitoring treatment response.
• The integrated tools facilitate efficient data labeling and marker analysis for pCD research.