Abstract
Accurate localization of 2D ultrasound planes within the 3D fetal brain is paramount for automated fetal assessment. However, freehand ultrasound acquisition is highly operator-dependent, and mapping 2D cross-sections to 3D anatomy remains a major challenge. Traditional pose estimation methods relying on Euler angles or quaternions suffer from discontinuities and singularities, leading to ambiguous localization and poor fetal brain recovery. We propose USPose, a novel SO(3)–equivariant deep learning framework that directly maps ultrasound features to the rotation manifold, capturing pose distributions rather than single-point estimates. By integrating a spatial-parameter-based regression network, USPose enables precise, real-time plane localization and reduces rotational ambiguity. Our approach achieves state-of-the-art localization accuracy and competitive resampled image quality compared with task-specific and foundation-model approaches. Furthermore, US-Pose yields better poses than prior work for downstream automated 2D-to-3D reconstruction on real clinical ultrasound data, thereby lowering the barrier to interpretation for quantitative volumetric analysis.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/ASMUS_026.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=9xQVUGRumT
BibTex
@InProceedings{RamJay_USPose_MICCAISAT2026,
author = { Ramesh, Jayroop AND Eid, Mark C. AND Yeung, Pak-Hei AND Henriques, Joao F. AND Rupprecht, Christian AND Namburete, Ana I. L.},
title = { { USPose: Improving 3D Fetal Brain Plane Localization for Volumetric Reconstruction from 2D Ultrasound with SO(3)-Equivariance } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17276},
month = {pending},
page = {pending}
}