Abstract
Automatic analysis of head and neck (H&N) cancer on fused FDG-PET/CT can make radiotherapy planning faster and more consis- tent. The HECKTOR 2026 challenge unites three clinical tasks on scans pooled from several hospitals, outlining the primary tumour and involved lymph nodes, assigning the tumour and nodal (TN) stage, and predicting recurrence-free survival. We present our solution to all three. Because a model accurate on the hospitals it trained on can degrade on an unseen one, we build every component for cross-centre stability. Tumour and node volumes are segmented with a self-configuring U-Net. Nodal stage is read directly from a few geometric measurements of the segmented nodes, chief among them a greatest-dimension diameter that mirrors the clinical size criterion, and refined by a fixed rule on metabolic activity. Tumour stage fuses a tabular classifier with a convolutional network on a tumour-centred crop, and recurrence-free survival is estimated with a Cox model over clinical variables and predicted stage. Every staging and survival component is chosen by its accuracy on a hospital held out during development. The pipeline attains a weighted validation score of 0.6467, and our two test-phase submissions score 0.5693 and 0.5733 on the hidden test set. In nodal staging the geometric head is best on both the average and the weakest centre, in survival no higher-capacity esti- mator beats a linear model, and in tumour staging an image model is stronger while a tabular anchor holds the weakest centre. Code is avail- able at https://github.com/vansh287/HECKTOR2026-VisionII. Team: Vision-II.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/HECKTOR_016.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=AYjiPhdkcv
BibTex
@InProceedings{PanVan_Geometry_MICCAISAT2026,
author = { Pandit, Vansh AND Shah, Ovais Iqbal AND Jakhetiya, Vinit AND Subudhi, Badri N.},
title = { { Geometry Meets Metabolism: Interpretable Segmentation, Staging, and Prognosis in Head and Neck PET/CT } },
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}
}
