Abstract

Automatic segmentation of the mitral valve can reduce manual work and make image analysis faster. The MICCAI 2026 Mitral Valve Anatomy Analysis (MVAA) challenge includes three segmentation tasks: cardiac computed tomography (CT), 3D transesophageal echocardiography (3D~TEE), and intraoperative surgical video. Only a small number of CT scans and surgical video frames are labeled, while many unlabeled images are available.

A separate model is trained for each task using the self-configuring nnU-Net framework. For labeled data, soft Dice loss and cross-entropy are combined with a multi-class boundary loss. The weight of the boundary loss is gradually increased during training. For cardiac CT and surgical video, unlabeled data are also used through FixMatch-style weak–strong consistency learning. A weak view created with geometric augmentation generates confident pseudo-labels. A strong view of the same image receives additional intensity augmentation and is trained to match the accepted pseudo-labels. The fully labeled 3D~TEE task uses only supervised training.

On the validation sets, the method achieved Dice similarity coefficients of 0.834 for cardiac CT, 0.789 for 3D~TEE, and 0.649 for surgical video. Performance was also evaluated using the 95th-percentile Hausdorff distance and the average surface distance.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/MVAA_089.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=GSzVMrvDxt

BibTex

@InProceedings{SinDur_SemiSupervised_MICCAISAT2026,
        author = { Singh, Durgesh Kumar AND Thomas, Sarina AND Cao, Qing AND Kampffmeyer, Michael C. AND Jenssen, Robert AND Solberg, Anne Schistad},
        title = { { Semi-Supervised nnU-Net with Weak–Strong Consistency and Boundary Refinement 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