Abstract

Accurate brain metastasis segmentation must balance overlap, boundary fidelity, and lesion detection, especially for small multifocal lesions and postoperative resection cavities. We present \method, the submitted frozen Docker inference pipeline for BraTS-METS 2026 Task~1. It combines a modality-aware state-space semantic expert and a residual-encoder region expert; their ET, RC, TC, and WT probabilities are fused by region, reconstructed hierarchically, and refined by RC gating and component verification. A conservative auxiliary lesion-recovery step is included only for rare case-level recovery rather than as a primary source of aggregate metric improvement.

In a development-set evaluator-aligned local analysis on the 202-scan, 122-patient split, BraTS-evaluation v0.0.8 yielded mean large-lesion DSC, large-lesion NSD, all-instance lesion-wise F1, and small-lesion F1 of 0.4704, 0.4908, 0.5841, and 0.2377, respectively. Patient-level cluster bootstrap showed wide RC uncertainty, including RC large-lesion DSC 0.1714 [0.0189, 0.3628]. These values are not Synapse validation-server results, because they were computed locally on our 202-scan development split. They should also not be interpreted as independent generalization estimates, because the same split was used for model selection, threshold tuning, OOF fitting of lightweight refinement modules, and diagnostic analysis.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/BraTS_METS_004.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=adMQR8qtDb

BibTex

@InProceedings{BiaYut_LesionAware_MICCAISAT2026,
        author = { Bian, Yutong AND Sun, Jindong},
        title = { { Lesion-Aware Hybrid State-Space and Residual-Encoder Fusion for Brain Metastasis Segmentation } },
        booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
        year = {2026},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 17253},
        month = {pending},
        page = {pending}
}


back to top