Abstract
Colorectal cancer (CRC) is the third most frequently diagnosed cancer and the second leading cause of cancer-related death worldwide. Accurate preoperative identification of T4 tumors is clinically important because it may influence treatment planning, including the use of neoadjuvant therapies. However, distinguishing T4 from non-T4 CRC on routine CT scans remains challenging in clinical practice. Previous deep learning approaches generally rely on manual tumor segmentation or predefined regions of interest. In this work, we instead investigate whether T4 tumors can be identified directly from whole CT scans without prior tumor localization. To evaluate this hypothesis, we trained two convolutional neural networks, ResNet-50 and Rad-ResNet-50, on 96 preoperative CT scans (171,531 axial slices) from our private database and applied six-fold cross-validation at the patient level. Slice-level predictions were aggregated by majority voting to obtain CT-level classifications. The best model achieved an AUC of 0.638 at the CT scan level. Although the predictive performance remains moderate, the results demonstrate that whole CT scans contain sufficient discriminative information for deep learning models to distinguish T4 from non-T4 CRC without prior segmentation. These findings support the feasibility of a fully automated staging approach and provide a baseline for future work using larger cohorts and more advanced architectures.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/CaPTion_026.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/profile?id=~Agust%C3%ADn_Ortiz_Lattuada1
BibTex
@InProceedings{OrtAgu_Deep_MICCAISAT2026,
author = { Ortiz Lattuada, Agustín AND Richard, Antoine T. AND Villeneuve, Laurent AND Gouttard, Sylvain AND Rousset, Pascal AND Kepenekian, Vahan AND Ladjal, Hamid},
title = { { Deep Learning-Based T4 Colorectal Cancer Staging from Preoperative CT Scans } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17278},
month = {pending},
page = {pending}
}