Abstract
We describe a vessel-guided submission to TopAneu 2026, covering vessel-wise aneurysm localization (Task 1) and location-labelled aneurysm segmentation (Task 2) in CTA and MRA. A shared vessel prior is selected from ten configurations derived from a native nnU-Net and released TopBrain 2025 models. On a 68-case internal evaluation subset, the selected, taxonomy-aligned UZH (re-labelled) model reaches 0.801 multiclass and 0.899 binary vessel Dice. A vessel-guided Task 1 classifier that corrects lateralized labels during mirror augmentation and uses rarity-aware sampling achieves a mean per-class precision/recall/MCC score of 0.195 on the primary split and 0.181 on an independent split; aneurysm copy-paste augmentation yields a 12\% relative improvement over its matched baseline. Task 2 experiments show that the initial unweighted presence loss collapses under severe voxel imbalance, whereas the revised foreground-weighted Dice+BCE loss yields mean Dice scores of 0.516 on CT and 0.469 on MR validation. These experiments identify taxonomy alignment and severe class imbalance as the main shared constraints.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/TopAneu_048.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=7tOzT5fxVO
BibTex
@InProceedings{TasFar_VesselGuided_MICCAISAT2026,
author = { Tasharofi, Farid AND Shoaei Taklimi, Mersad AND Rist, Leonhard AND Maier, Andreas},
title = { { Vessel-Guided Aneurysm Localization and Segmentation for TopAneu 2026 } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17255},
month = {pending},
page = {pending}
}