Abstract
Pathologic complete response (pCR) after neoadjuvant treatment is an important endpoint in rectal cancer because it is closely associated with treatment response assessment, organ-preserving strategies, and individualized surgical planning. Accurate preoperative pCR prediction remains challenging for small, highly imbalanced multimodal cohorts. In our cohort, only 8 of 51 patients achieved pCR.
We propose a representation-aware multimodal integration framework that preserves complementary patient representations instead of directly concatenating heterogeneous modalities. The representations are independently modeled and integrated through a training-calibrated percentile-rank aggregation strategy, allowing heterogeneous model outputs to be combined without learning ensemble weights.
In 5-repeat 4-fold stratified cross-validation, the proposed framework achieved a fold-level AUC of 0.818 and an AUPRC of 0.585. Patient-level repeated out-of-fold evaluation achieved an AUC of 0.881 and an AUPRC of 0.614, with a bootstrap 95% confidence interval of 0.778–0.973. These results suggest that representation-aware multimodal integration is an effective strategy for small-sample, imbalanced pCR prediction.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/MWM_094.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=eq2mEPMcG2
BibTex
@InProceedings{LeePin_RepresentationAware_MICCAISAT2026,
author = { Lee, Ping-Lun AND Kao, Yun-Ching AND Lin, Cheng-Kuan AND Tseng, Yu-Chee AND Huang, Eng-Yen AND Huang, Tzu-Ting AND Huang, Kuan-Ching},
title = { { Representation-Aware Multimodal Integration for Small-Sample pCR Prediction in Rectal Cancer } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17262},
month = {pending},
page = {pending}
}