Seminar: Graduate Seminar
Segmentation-Guided Region-Based Classification of Crohn’s Disease Imaging Findings from Multi-Contrast MRE
Date:
October,06,2026
Start Time:
13:00 - 14:00
Location:
1061, Meyer Building
Zoom:
Zoom link
Add to:
Lecturer:
Sarah Teitz
Research Areas:
| Crohn’s disease (CD) is a chronic inflammatory bowel disease that can cause progressive bowel damage and complications. Magnetic resonance enterography (MRE) enables disease monitoring through non-invasive evaluation of the entire intestinal tract and surrounding extramural tissues; however, its interpretation is complex, requiring the assessment of multiple radiological findings across anatomical regions and imaging contrasts. Existing automated classification methods typically rely on manual spatial annotations, or provide only coarse disease-level classification, and generally use a single MRE contrast. To address these limitations, a fully automated framework was developed for region-based, multi-label classification of eight CD-related radiological findings from T1- and T2-weighted MRE. The framework uses a pretrained 3D nnU-Net to segment the ileum, contrast-specific feature extractors to obtain multi-resolution features, and segmentation-guided masked Generalized Mean pooling to focus feature aggregation on anatomically relevant regions. The pooled T1- and T2-derived features are combined using either linear, learned, or attention-based modality weighting. Experiments comparing single- and multi-contrast classification, modality-weighting strategies, and whole-image versus segmentation-guided pooling showed that segmentation guidance enables more precise classification and that multi-contrast imaging with attention-based weighting achieves the best performance. |
| M.Sc. student under the supervision of Associate Prof. Moti Freiman and Prof. Israel Cohen.
|