UBone3D: Physics-Rectified Conditional Flow Matching for Anatomical 3D Shape Completion from Ultrasound

2026-09-10

AI Summary

本文提出UBone3D框架，通过物理校正的条件流匹配方法，解决从超声点云中恢复完整解剖结构的问题。

Abstract

Three-dimensional ultrasound (US) is a safe, radiation-free complementary modality to CT and X-rays for longitudinal monitoring, yet its segmentation-derived partial point clouds are extremely artifact-laden. Consequently, it is challenging to recover a clean and complete anatomical structure from such US point clouds. In this paper, we present UBone3D, a novel framework based on physics-rectified conditional flow matching (CFM) that performs point cloud completion directly from partial US observations. UBone3D models deterministic physics artifacts (e.g., surface thickening, streaking, dropouts) via a simulated physics proxy, and introduces test-time physics rectification to steer the shape completion. At inference, the completion is jointly steered by two decoupled forces: (1) anatomical plausibility enforced by a CT-trained generative shape prior, BoneFM, and (2) physics consistency enforced by USimNet in the ultrasound formation space. Extensive experiments on simulated and in-vivo data demonstrate significant improvements in reconstruction accuracy and anatomical fidelity over existing baselines.

Problem

• 3D ultrasound
• point cloud completion
• anatomical structure
• artifact-laden

Innovation

• Physics-rectified Conditional Flow Matching
• Anatomical 3D Shape Completion
• Ultrasound Point Clouds
• Simulated Physics Proxy
• Test-time Physics Rectification