Abstract
We describe our submission to the three HECKTOR 2026 subtasks: segmentation of primary tumors and involved lymph nodes, T- and N-stage classification, and recurrence-free survival (RFS) prediction from paired PET/CT and clinical variables. The submission integrates established segmentation and tabular models in a fixed, task-specific pipeline. An ensemble of five nnU-Net cascade models predicts the pri- mary gross tumor volume (GTVp) and gross tumor volume of involved cervical lymph nodes (GTVn) over a wide head-and-neck field of view. A local refinement network then uses CT, PET, and the predicted GTVp and GTVn masks as input channels. A predefined fallback handles empty or near-empty segmentation outputs. The T- and N-stage classifiers com- bine clinical variables with handcrafted imaging features derived from the refined masks. The RFS model combines an ensemble of extremely ran- domized trees trained on observed recurrence status with an extremely randomized survival forest. Segmentation predictions are averaged across five-fold models with mirroring, whereas tabular predictions are averaged across five random seeds. No external data or externally pretrained mod- els are used. Official validation scores were Mean Dice 0.6394, T- and N-stage balanced accuracies 0.4931 and 0.5184 (mean 0.5058), and RFS C-index 0.8286; corresponding test scores were 0.6498, 0.5896, 0.5908, and 0.6593, respectively. Grand Challenge team: IMPACT lab.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/HECKTOR_006.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=AX8pwjnECf
BibTex
@InProceedings{XuChe_WideField_MICCAISAT2026,
author = { Xu, Chenfan AND Wang, Hao AND Liu, Zhentao AND Luo, Tao AND Luo, Lei AND Hu, Han AND Du, Wen AND Cui, Zhiming},
title = { { Wide-Field PET/CT Segmentation, Local Refinement, and Clinical-Variable Models for HECKTOR 2026 } },
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}
}