Abstract
Segmentation, staging, and prognosis of head and neck can- cer are traditionally treated as separate tasks, despite their clinical inter- dependence. We present an end-to-end inference pipeline for the HECKTOR 2026 challenge that jointly predicts primary tumor and nodal seg- mentation (GTVp/GTVn), TN stage, and recurrence-free survival from [18F]FDG-PET/CT and clinical data spanning multiple centers, devel- oped on 8 centers and evaluated on centers held out during development. Staging and prognosis combine deep learning and radiomics branches through a diversity-driven ensemble strategy: candidate sources are screened by out-of-fold strength and Spearman rank correlation, favoring complementary, cross-paradigm information over marginal gains in individual source strength, with the final ensemble validated by paired bootstrap testing. Both branches consistently use Task 1’s predicted, rather than ground-truth, masks, matching training time and inference-time conditions. Model selection throughout relied on leave-one-center-out cross-validation, prioritizing honest cross-center generalization, though Task 1’s five-fold splits were not enforced to be center-disjoint, over same-distribution point estimates. Two submitted configurations, differing only in their prognosis ensemble, achieved weighted scores of 0.7205 and 0.6720 on the official Validation Phase leaderboard; the primary configuration’s advantage was confirmed on a larger internal out-of-fold cohort (n=676) with statistical significance (p ≈ 0.006).
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/HECKTOR_013.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=0ApAYywpCc
BibTex
@InProceedings{OuyJin_Predicted_MICCAISAT2026,
author = { Ouyang, Jin AND Ding, Mianyong AND Smeets, Esther AND Ma, Baoqiang},
title = { { Predicted Masks Throughout: A Cross-Center Generalizable Pipeline for Joint Head and Neck Tumor Segmentation, Staging, and Prognosis } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17271},
month = {pending},
page = {pending}
}