Abstract
Accurate landmark detection is essential for ultrasound biometry but remains challenging due to limited annotations and substantial anatomical variability. In this work, we propose LandWorld, a latent predictive foundation model for universal landmark-based ultrasound biometry. Our contribution is three-fold. First, LandWorld adopts UltraFedFM as a shared encoder to capture target-agnostic ultrasound representations and attaches lightweight multi-task heads to accommodate target-independent predictions. Second, to exploit abundant unlabeled images, we introduce dual-consistency regularization into the semi-supervised pipeline to stabilize target-specific predictions and improve domain-balanced representation learning. Last, we develop a landmark-aware world modeling strategy based on feature prediction in the latent space, further improving anatomical context extraction and landmark understanding. Extensive experiments on 34,658 labeled/197,102 unlabeled ultrasound images with 4 clinical domains, 9 targets, 60 landmarks, and 28 biometric parameters, LandWorld shows better performance than existing foundation models and the leading challenge entries. Codes and models will be publicly released upon acceptance.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/FoundUS_068.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=NIk9rvvcKc
BibTex
@InProceedings{DuaYao_LandWorld_MICCAISAT2026,
author = { Duan, Yaofei AND Huang, Yuhao AND Chen, Rusi AND Huang, Sihan AND Zhang, Hongyuan AND Tan, Tao},
title = { { LandWorld: A Latent Predictive Foundation Model for Landmark-Based Ultrasound Biometry } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17261},
month = {pending},
page = {pending}
}