Automatic Localization and Classification of Crohn's Disease Activity in Computed Tomography Enterography Images...

Area of Science:

• Artificial Intelligence in Medical Imaging
• Gastroenterology
• Radiology

Background:

• Accurate localization and grading of Crohn's disease (CD) lesions on computed tomography enterography (CTE) are crucial for efficient diagnosis.
• Challenges in CD lesion detection include variable sizes, shapes, and blurred boundaries on CTE images.

Purpose of the Study:

• To develop an automated detection and classification model for CD lesions using a one-stage YOLOv5-based framework (CD-YOLO).
• To enhance feature extraction and multi-scale feature fusion for improved detection of CD lesions with varying characteristics.

Main Methods:

• Development of CD-YOLO, a one-stage model based on YOLOv5, incorporating specialized modules for feature extraction and multi-scale fusion.
• Retrospective collection and analysis of CTE images from 233 pathologically and endoscopically confirmed CD patients.
• Performance evaluation using mAP@0.5, precision, recall, and visualization via Gradient-weighted Class Activation Mapping (Grad-CAM).

Main Results:

• CD-YOLO achieved mAP@0.5 scores of 92.1% for active CD and 84.2% for remission CD.
• The model demonstrated significant improvements over YOLOv5s, with mAP@0.5 increasing by 2.0% for active and 4.9% for remission CD.
• Missed detection and misdetection rates for CD lesions were reduced by 4.9% and 4.18%, respectively.

Conclusions:

• The CD-YOLO framework shows superior performance in localizing and classifying CD lesion activity on CTE.
• CD-YOLO has the potential to serve as an effective computer-aided diagnostic tool for radiologists and gastroenterologists.
• The model can aid in clinical teaching and improve the overall diagnostic workflow for Crohn's disease.