Abstract

We present our submission to the HECKTOR 2026 challenge, which addresses three interconnected tasks on multimodal PET/CT of head and neck cancer patients: gross tumor volume (GTVp/GTVn) segmentation, TN staging classification, and recurrence-free survival (RFS) prognosis. Our pipeline shares a single preprocessing stage across all three tasks – PET/CT resampling to a common isotropic grid, an intensity-driven localization of the head-and-neck region, and a fixed-size crop around it – followed by task-specific models. Segmentation is performed with a 5-fold nnU-Net (ResEnc UNet L, 3d_fullres) ensemble. TN staging and prognosis are each addressed with a dual-branch 3D convolutional network that fuses image/mask features with a clinical and radiomics feature vector, using two separate task-specific clinical feature definitions and normal- ization statistics fitted at training time. On the validation phase, our approach obtained a mean Dice score of 0.6194, a T-stage balanced accuracy of 0.4827, an N-stage balanced accuracy of 0.6721, and an RFS concordance index of 0.6964, yielding an overall weighted score of 0.6355 under the challenge’s official ranking formula. Corresponding testing-phase figures, to be provided by the organizers ahead of the camera-ready deadline, will be added in the final version. We dis- cuss design choices, in particular the use of task-specific radiomics connectivity conventions and a sign-corrected risk score for the prognosis head, and analyze the relative strengths and weaknesses of the three components.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/HECKTOR_014.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=mEFJmRdvLa

BibTex

@InProceedings{AisIkr_ADualBranch_MICCAISAT2026,
        author = { Aissiou, Ikram AND Fellah, Ahmed Riadh AND Guessoum, Sam AND Boukhiar, Naima},
        title = { { A Dual-Branch Fusion Pipeline for Head and Neck Tumor Segmentation, TN Staging, and Recurrence-Free Survival Prediction: 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}
}


back to top