Abstract
Bowel wall thickening (BWT) on CT occurs in both colon cancer and diverticulitis, two conditions requiring fundamentally different treatment. Their overlapping imaging appearance makes reliable differentiation difficult, particularly in emergency settings where rapid triage is critical. Existing deep learning approaches address this problem only partially, either targeting cancer-specific segmentation or classifying pathology within manually defined regions of interest. We present the first fully automated end-to-end framework combining 3D nnU-Net lesion localization with dual-channel 3D ResNet-18 classification. The pipeline achieved a malignancy sensitivity of 0.98 and specificity of 0.81 on the internal test set. External validation supported generalization, reaching up to 0.98 accuracy on public cancer datasets and maintaining a malignancy sensitivity of 0.74 under severe distribution shift in a heterogeneous multi-scanner Charité cohort, where four 3D baselines performed near chance. Input masking experiments showed radiologically plausible decision patterns, while segmentation failure analysis suggested that model–reference disagreements mainly reflect annotation incompleteness and boundary ambiguity.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/EMERGE_004.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
Open Review Page: https://openreview.net/forum?id=OBq3eOm2NA
BibTex
@InProceedings{BenZei_Fully_MICCAISAT2026,
author = { Ben Chaaben, Zeineb AND Schnabel, Julia A. AND Adams, Lisa C. AND Zirn, Christopher AND Lemke, Tristan AND Häntze, Hartmut AND Schäfer, Patricia AND Auer, Timo AND Bressem, Keno AND Ziegelmayer, Sebastian},
title = { { Fully Automated CT-Based Differential Diagnosis of Bowel Wall Thickening } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17266},
month = {pending},
page = {pending}
}