Abstract
Accurate assessment of tumor burden, disease staging at di- agnosis, and treatment outcome prediction is critical for personalized management of head and neck cancer. Recent advances in artificial intelli- gence have shown promise in complementing traditional prognostic models. However, clinical adoption is limited by the lack of standardized bench- marks and unified end-to-end solutions. In this work, we present a unified framework for joint tumor segmentation, TN staging, and recurrence-free survival prediction from 18F-FDG PET/CT imaging and clinical data. A 3D residual U-Net was first trained to segment primary tumors and adenopathies, providing a shared imaging backbone from which patient- level feature sets were derived for TN staging and outcome prediction. To improve precision in detecting metastatic lymph node involvement, nodal-specific optimization strategies were incorporated during training. Imaging and clinical features were subsequently combined through a late-fusion strategy, leveraging their complementary information to improve both staging and survival prediction. The proposed framework was evaluated as part of the MICCAI HECKTOR 2026 challenge.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/HECKTOR_009.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=1nFSzrNz5o
BibTex
@InProceedings{SzaJan_ALateFusion_MICCAISAT2026,
author = { Szalma, Janos AND Juhasz, Adam AND Schaerer, Joel AND Ebert, Emily AND Scarimbolo, Robert AND Bansal, Anu AND Delmonte, Alessandro},
title = { { A Late-Fusion Framework for Joint Tumor Segmentation, Staging, and Recurrence Prediction in Head and Neck Cancer } },
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}
}