DreamReg: Belief-Driven World Model for 2D-3D Ultrasound Registration

2026-06-17 · Source: Computer Vision and Pattern Recognition · Field: Health & Wellbeing — Medical Devices & Health Technology, Clinical Care & Medical Practice · Depth: Expert, quick

Summary

DreamReg is a novel belief-driven world-model framework designed to address the challenges of real-time 2D-3D ultrasound registration for surgical navigation. Existing methods struggle with partial observability, speckle noise, and action-dependent ultrasound acquisition, often being limited to one-shot or short-horizon operations. DreamReg formulates registration as a continuous belief updating process over rigid transformations, maintaining a latent belief state that integrates past observations and pose information. It refines transformations through learned dynamics as new ultrasound slices arrive. During training, DreamReg learns from probe-motion trajectories mimicking clinical scanning, conditioning pose refinement on current observations. For inference, it employs internal imagination, simulating candidate probe motions and their predicted observations to converge on an accurate rigid transformation. Experiments on the CAMUS and u-RegPro datasets demonstrate DreamReg's improved robustness and competitive registration accuracy compared to state-of-the-art methods.

Key takeaway

For Machine Learning Engineers developing medical imaging solutions, DreamReg demonstrates that integrating belief-driven world models can significantly improve real-time 2D-3D ultrasound registration. You should consider adopting continuous belief updating and internal imagination techniques to enhance robustness against partial observability and speckle noise. This approach offers a path to more accurate and reliable surgical navigation systems, especially when dealing with action-dependent data acquisition.

Key insights

DreamReg uses a belief-driven world model and internal imagination for robust 2D-3D ultrasound registration, overcoming real-time challenges.

Principles

• Continuous belief updating enhances registration robustness.
• World models can simulate future states for refinement.
• Mimicking clinical behavior improves model training.

Method

DreamReg formulates 2D-3D registration as belief updating. It maintains a latent belief state, refines transformations via learned dynamics, and uses internal imagination to simulate probe motions and observations for convergence.

In practice

• Use belief-driven models for sequential data fusion.
• Integrate simulated outcomes for robust pose estimation.
• Train models with realistic action-dependent trajectories.

Topics

• 2D-3D Registration
• Ultrasound Imaging
• World Models
• Belief Updating
• Surgical Navigation
• Medical Imaging

Best for: AI Scientist, Research Scientist, Machine Learning Engineer

Related on AIssential

• Look-Closer-Then-Diagnose: Confidence-Aware Ultrasound VQA via Active Zooming — A novel VLM framework integrates lesion-focused reasoning and uncertainty-aware rewards for improved ultrasound diagnosis.
• Transformation-driven generation of comparable projection images from multimodal anatomical scenes — A transformation-driven framework generates comparable synthetic projection images from multimodal anatomical scenes by explicitly modeling observation processes.
• Towards World Models in Biomedical Research — Biomedical world models simulate biological futures by learning intervention-conditioned dynamics from multiscale latent states.
• Rethinking Patient Education as Multi-turn Multi-modal Interaction — MedImageEdu benchmarks multi-turn, multimodal patient education, revealing gaps in visual grounding, safety, and handling emotional interactions.
• SMART: A Flexible, Interpretable, and Scalable Spatio-temporal Brain Atlas from High-Resolution Imaging Data — SMART creates flexible, interpretable brain atlases by decoupling global disease dynamics from patient-specific anatomy using differential equations and Neural Cellular Automata.
• A Koopman-Bayesian Framework for High-Fidelity, Perceptually Optimized Haptic Surgical Simulation — A Koopman-Bayesian framework improves haptic surgical simulation by unifying nonlinear dynamics and perceptual psychophysics.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computer Vision and Pattern Recognition.