Real-Time Novel-View Freehand Ultrasound Imaging via Point-Cloud Rendering and Diffusion-Bridge Completion

Hanrui Shi, Boris Mailhé, Zheyuan Zhang, Yikang Liu, Xiao Chen, Ankush Mukherjee, Terrence Chen, Shanhui Sun
Proceedings of The 9th International Conference on Medical Imaging with Deep Learning, PMLR 315:2283-2296, 2026.

Abstract

Freehand ultrasound imaging is limited by sparse sampling and restricted probe coverage, which prevent consistent visualization of unseen planes and oblique orientations. We propose a real-time framework for novel-view ultrasound imaging that combines point-cloud rendering with diffusion-bridge completion. Given a sequence of 2D B-mode images and tracked probe poses, each novel view is first rendered as a partially observed slice from the reconstructed point cloud geometry, then completed by an Image-to-Image Schr{ö}dinger Bridge (I$^2$SB) model to synthesize anatomically coherent textures. The diffusion-bridge formulation accelerates convergence by conditioning on visible regions instead of noise, enabling stochastic yet efficient generation. A latent I$^2$SB variant further improves computational efficiency for high-resolution ultrasound data. Experiments on an abdominal dataset demonstrate realistic novel-view synthesis with fine structural continuity and real-time inference ($<$0.2 seconds per view), outperforming standard diffusion inpainting baselines in both speed and visual fidelity. The proposed method provides an efficient generative approach for interactive and view-adaptive ultrasound visualization.

Cite this Paper

BibTeX
@InProceedings{pmlr-v315-shi26b, title = {Real-Time Novel-View Freehand Ultrasound Imaging via Point-Cloud Rendering and Diffusion-Bridge Completion}, author = {Shi, Hanrui and Mailh{\'e}, Boris and Zhang, Zheyuan and Liu, Yikang and Chen, Xiao and Mukherjee, Ankush and Chen, Terrence and Sun, Shanhui}, booktitle = {Proceedings of The 9th International Conference on Medical Imaging with Deep Learning}, pages = {2283--2296}, year = {2026}, editor = {Huo, Yuankai and Gao, Mingchen and Kuo, Chang-Fu and Jin, Yueming and Deng, Ruining}, volume = {315}, series = {Proceedings of Machine Learning Research}, month = {08--10 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v315/main/assets/shi26b/shi26b.pdf}, url = {https://proceedings.mlr.press/v315/shi26b.html}, abstract = {Freehand ultrasound imaging is limited by sparse sampling and restricted probe coverage, which prevent consistent visualization of unseen planes and oblique orientations. We propose a real-time framework for novel-view ultrasound imaging that combines point-cloud rendering with diffusion-bridge completion. Given a sequence of 2D B-mode images and tracked probe poses, each novel view is first rendered as a partially observed slice from the reconstructed point cloud geometry, then completed by an Image-to-Image Schr{ö}dinger Bridge (I$^2$SB) model to synthesize anatomically coherent textures. The diffusion-bridge formulation accelerates convergence by conditioning on visible regions instead of noise, enabling stochastic yet efficient generation. A latent I$^2$SB variant further improves computational efficiency for high-resolution ultrasound data. Experiments on an abdominal dataset demonstrate realistic novel-view synthesis with fine structural continuity and real-time inference ($<$0.2 seconds per view), outperforming standard diffusion inpainting baselines in both speed and visual fidelity. The proposed method provides an efficient generative approach for interactive and view-adaptive ultrasound visualization.} }
Endnote
%0 Conference Paper %T Real-Time Novel-View Freehand Ultrasound Imaging via Point-Cloud Rendering and Diffusion-Bridge Completion %A Hanrui Shi %A Boris Mailhé %A Zheyuan Zhang %A Yikang Liu %A Xiao Chen %A Ankush Mukherjee %A Terrence Chen %A Shanhui Sun %B Proceedings of The 9th International Conference on Medical Imaging with Deep Learning %C Proceedings of Machine Learning Research %D 2026 %E Yuankai Huo %E Mingchen Gao %E Chang-Fu Kuo %E Yueming Jin %E Ruining Deng %F pmlr-v315-shi26b %I PMLR %P 2283--2296 %U https://proceedings.mlr.press/v315/shi26b.html %V 315 %X Freehand ultrasound imaging is limited by sparse sampling and restricted probe coverage, which prevent consistent visualization of unseen planes and oblique orientations. We propose a real-time framework for novel-view ultrasound imaging that combines point-cloud rendering with diffusion-bridge completion. Given a sequence of 2D B-mode images and tracked probe poses, each novel view is first rendered as a partially observed slice from the reconstructed point cloud geometry, then completed by an Image-to-Image Schr{ö}dinger Bridge (I$^2$SB) model to synthesize anatomically coherent textures. The diffusion-bridge formulation accelerates convergence by conditioning on visible regions instead of noise, enabling stochastic yet efficient generation. A latent I$^2$SB variant further improves computational efficiency for high-resolution ultrasound data. Experiments on an abdominal dataset demonstrate realistic novel-view synthesis with fine structural continuity and real-time inference ($<$0.2 seconds per view), outperforming standard diffusion inpainting baselines in both speed and visual fidelity. The proposed method provides an efficient generative approach for interactive and view-adaptive ultrasound visualization.
APA
Shi, H., Mailhé, B., Zhang, Z., Liu, Y., Chen, X., Mukherjee, A., Chen, T. & Sun, S.. (2026). Real-Time Novel-View Freehand Ultrasound Imaging via Point-Cloud Rendering and Diffusion-Bridge Completion. Proceedings of The 9th International Conference on Medical Imaging with Deep Learning, in Proceedings of Machine Learning Research 315:2283-2296 Available from https://proceedings.mlr.press/v315/shi26b.html.

Related Material