Local Assessment and Small Bowel Crohn's Disease Severity Scoring using AI.

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Imaging

Background:

• Current Crohn's disease (CD) severity assessment relies on manual measurements and qualitative evaluations of CT enterography (CTE) images, which have limitations.
• The complex anatomy and disease distribution in the small bowel pose challenges for accurate and standardized scoring.

Purpose of the Study:

• To develop and validate a machine learning and computer vision approach for localized, automated, and standardized scoring of CD severity in the small bowel.
• To overcome the limitations of manual assessments in CTE imaging.

Main Methods:

• A hybrid model combining deep learning, 3-D Convolutional Neural Networks (CNNs), and Random Forest was developed.
• Two radiologists evaluated disease severity at 7.5 mm intervals on 236 CTE scans of the distal and terminal ileum.
• Model performance was assessed using precision, sensitivity, Cohen's score, and accuracy on a hold-out test set.

Main Results:

• The hybrid model achieved precision and sensitivity between 42.4% and 84.1% across different severity categories.
• The model's Cohen's score (κ = 0.83) and accuracy (70.7%) were comparable to inter-observer agreement among radiologists (κ = 0.87, accuracy = 76.3%).
• The model accurately predicted disease length (r = 0.83) and identified moderate-to-severe disease portions with 91.51% accuracy.

Conclusions:

• The automated hybrid model provides a standardized, reproducible, and quantitative local assessment of small bowel CD severity.
• This AI-driven approach demonstrates significant value in enhancing CD severity assessment and management.