Abstract
Clinical ultrasound biometry spans prenatal, intrapartum, and cardiac imaging, yet models are typically built per task, return measurements with no notion of their own reliability, and silently de- grade under real multi-vendor conditions. We recast biometry as a step toward a medical world model : a single self-supervised encoder builds a unified anatomical state (landmarks → 20+ clinical parameters) across all three regimes; a generative statistical shape model provides a population-level anatomical prior; and an ensemble-disagreement signal yields a per-case self-assessment that flags unreliable measurements. Crucially, through controlled experiments on real ground truth we di- agnose that the detector’s poor generalization is caused by geometric fragility: trained without keypoint-space augmentation, it collapses un- der the slightly different probe geometry of the test distribution (a 10% shift inflates error 2×). Routing landmarks through the augmentation makes the model geometrically invariant and, unlike every architecture change we tried, transfers to the official test set; with a high-resolution branch and per-view selection this advances our unified model from mean radial error 27.8 to 23.0, from rank 27 to rank 4 on the MICCAI 2026 FU Biometry leaderboard, while the self-assessment recovers 62% of the worst cases at a 20% review budget. We report ablations, data- quality corrections, and negative results honestly.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/FoundUS_061.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=EY7JMVa4cd
BibTex
@InProceedings{SinGag_SelfAssessing_MICCAISAT2026,
author = { Singh, Gagneet AND Dukre, Adinath Madhavrao AND Razzak, Imran},
title = { { Self-Assessing Anatomical World Model for Unified Multi-Domain 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}
}