Abstract
Head and neck cancer management relies on fluorodeoxyglu- cose positron emission tomography / computed tomography (FDG- PET/CT) to delineate the primary and nodal gross tumor volumes (GTVp/GTVn), assign the tumor and nodal (TN) stage, and estimate prognosis; HECKTOR 2026 casts these steps as one pipeline of segmentation, T/N staging, and recurrence-free survival (RFS) prediction. We present a three-stage, leakage-aware cascade whose downstream models consume only fold-aligned out-of-fold (OOF) upstream predictions. Its core, the Efficient SwiGLU Residual Encoder U-Net, modernizes nnU-Net v2 ResEnc-M by re-examining each borrowed component under controlled ablations: mean Dice improves from 0.6759 to 0.6948 while total parameters fall by 19% and decoder parameters by 94.5%. The submitted model attains a pooled OOF mean Dice of 0.6957, significantly above a final-setting ResEnc-M baseline; T/N balanced accuracies are 0.5662 and 0.5663, and the RFS ensemble reaches a Harrell C-index of 0.6762, versus 0.5982 for a clinical-only baseline. On the official test set, the weighted score is 0.5776 (Dice 0.6186; T/N 0.4107/0.5708; C-index 0.6280). Our results favor task-specific capacity reallocation over direct transfer of general-purpose design choices. Team: NCKU Tingyi Solo.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/HECKTOR_002.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=NJRyWWsMe2
BibTex
@InProceedings{ChaTin_Modernizing_MICCAISAT2026,
author = { Chang, Ting-Yi},
title = { { Modernizing nnU-Net with Convolutional SwiGLU for Segmentation, Staging, and Prognosis in 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}
}
