Abstract

Region- and surface-oriented objectives yield complementary error profiles in 3D transesophageal echocardiography (TEE) segmentation, but the resulting experts are not functionally interchangeable. ReGate-MV combines region and surface-specialized experts through a spatial authority field and applies classwise largest connected component (LCC) repair after discretization. The field supports both equal global composition and a prediction-derived selective variant. On two organizer-defined internal validation splits, equal composition with repair, relative to the region baseline, increased the Dice similarity coefficient (DSC) by 0.0041 and 0.0029, reduced Hausdorff distance (HD) by 19.3% and 15.4%, and reduced average surface distance (ASD) by 6.9% and 15.6%, respectively. Relative to a repaired control comprising two region experts, the complete surface-specialized pathway yielded higher DSC and lower ASD on both splits. Ablations showed that probability composition improved overlap and reduced ASD, whereas LCC reduced HD in every matched comparison. In post-inference analysis, 84.5% of substitutions made by the selective rule lay within two voxels of the ground-truth boundary, although ground truth was not used to construct the gate. In the official hidden evaluation, replacing only the TEE chain of the organizer reference pipeline increased DSC by 0.0902 and reduced HD and ASD by 3.5366 and 0.4134, respectively; this comparison reflects full-pipeline replacement rather than the isolated effect of expert composition. Independent CT and video branches complete the three-task submission under task-specific rules.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/MVAA_082.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=0JNQH0ZkvC

BibTex

@InProceedings{LiYin_ReGateMV_MICCAISAT2026,
        author = { Li, Yingfa AND Chen, Yijie AND Zhu, Jiawei AND Huang, Junlin AND Sun, Entao AND Liang, Shuang AND Gu, Yu},
        title = { { ReGate-MV: Role-Aware Region–Surface Expert Composition with Structural Repair for Mitral Valve Segmentation } },
        booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
        year = {2026},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 17268},
        month = {pending},
        page = {pending}
}


back to top