Detection of Inflammatory Bowel Diseases in MRE Images Using Multi-Slice Multi-Label Spatial Context Learning

Crohn’s disease is a chronic inflammatory bowel disease, and accurate detection of inflammatory regions in MRE images is essential for its diagnosis and severity assessment. Traditional single-slice-based MRE analysis fails to capture the spatial context across consecutive slices, limiting detection performance. This study proposes a Multi-Slice Multi-Label (MSML) method that integrates spatial context into the YOLOv4 model. By assigning labels based on the occurrence frequency of inflammatory regions across three consecutive slices, the method enhances spatial awareness. Experimental results demonstrate that the proposed MSML approach achieves higher mAP (69.27%) and improved detection consistency compared to single-slice-based models, while also reducing false positive detections.