Abstract
Multimodal recurrence prediction in oncology increasingly combines clinical variables, radiomics and learned image representations, but fair comparison remains dicult when studies use dierent preprocessing, feature-selection, modelling and validation strategies. We introduce a reproducible benchmark for standardized comparison of clinical data, CT radiomics and CT foundation-model embeddings in cancer recurrence prediction. RADCURE dataset was used as the main multimodal cohort for 5-year recurrence-free survival prediction, including 1407 aligned patients across all modalities. A xed 80/20 split assigned 1125 patients to development and 282 to holdout testing, with preprocessing, feature selection, sampling, model selection, hyperparameter tuning and threshold optimization restricted to the development set. HECKTOR cancer dataset was included to assess framework portability. In the RADCURE holdout evaluation, clinical variables achieved the strongest performance in terms of balanced accuracy, while late probability fusion of multimodal features achieved the highest ROC-AUC. CT radiomics and CT embeddings show more moderate unimodal performance. Overall, the benchmark provides a transparent and reproducible framework for comparing multimodal recurrence prediction pipelines before clinical translation. The benchmark code is publicly available on GitHub Link.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/MultiTab_021.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=cs9JqiF0UD
BibTex
@InProceedings{TerMan_AReproducible_MICCAISAT2026,
author = { Terradillos Perea, Manuel AND Guetarni, Bilel AND Rodríguez González, Blanca AND Rodríguez Vila, Borja AND Torrado-Carvajal, Ángel AND Malpica, Norberto AND Benhabiles, Halim},
title = { { A Reproducible Public Benchmark for End-to-End Machine Learning from Multimodal Tabular Data in Cancer Recurrence Prediction } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17263},
month = {pending},
page = {pending}
}