Abstract

Head and neck (H&N) cancer is the 7th most common malignancy worldwide, with 5-year survival ranging between 25%-60% and treatment choice depending on disease stage - largely determined by tumor size and nodal spread. To this end, the MICCAI 2026 HEad and NeCK TumOR (HECKTOR) challenge focused on i) the segmentation of primary tumors and lymph nodes, ii) their staging, and iii) prediction of recurrence-free survival (RFS). Here, we introduce GASP as an inte- grated framework to address all 3 tasks based on geometry-aware mod- eling. GASP was trained on multi-institutional (n=8) co-registered PET and CT scans, expert tumor and nodal annotations, and clinical records from 782 patients. Performance was evaluated on a patient-weighted leave-one-center-out cross validation, and on the official hold out test- ing cohort of 400 patients. Official testing results yield i) mean Dice of 0.652 for segmentation, ii) concordance index of 0.558 for RFS, and iii) balanced accuracies of 0.461 & 0.564 for tumor & nodal (TN) staging, respectively. GASP achieved an overall score of 0.5654, following HECK- TOR’s aggregated overall score with weightings of 0.25 for segmentation, 0.35 for TN staging, and 0.40 for survival. The best-performing configuration combined an nnU-Net segmentation ensemble with a label-audited and a size-focused logistic classifier for tumor and nodal staging, respectively, and a Cox proportional hazards model for survival. Code and trained weights are available on GitHub and Zenodo.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/HECKTOR_017.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=DRNzaZ5QML

BibTex

@InProceedings{KacSan_GASP_MICCAISAT2026,
        author = { Kachole, Sanket AND Bakas, Spyridon},
        title = { { GASP: Multimodal Geometry-Aware Segmentation, Staging, and Prognostication of 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}
}


back to top