Abstract
Automated ultrasound biometry is essential for fetal growth monitoring and cardiac function assessment, yet manual landmark an notation in low contrast images is time consuming and prone to inter observer variability. The MICCAI 2026 Medical World Model (MWM) Challenge promotes unified ultrasound analysis through a multicenter, multitask benchmark. However, foundation models face three key chal lenges: task heterogeneity, ambiguous tissue boundaries, and imbalanced data with noisy pseudo labels. We address these challenges with a uni fied framework comprising three components. First, a residual adapter conditioned on task identity specializes shared representations with mini mal additional parameters. Second, a progressive anchor region shrinking algorithm uses geometric consistency to refine measurement endpoints. Third, a task balanced hybrid semisupervised learning strategy improves the use of unlabeled multicenter data. On the MWM 2026 validation set, our method achieved an average MAE of 25.98, outperforming the baseline by 3.53 mm. It also reduced large errors in highly variable api cal four chamber views and images with ambiguous femoral boundaries. These results demonstrate the accuracy, robustness, and scalability of our framework for unified ultrasound biometry, with the potential to reduce clinical workload and improve measurement consistency across centers
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/FoundUS_111.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=JkNUGT1ZNk
BibTex
@InProceedings{SonShu_PULSEFM_MICCAISAT2026,
author = { Song, Shuru AND Chi, Jianning},
title = { { PULSE-FM: A Foundation Model for Ultrasound Biometry with Task Adaptation and Progressive Landmark Refinement } },
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}
}