Abstract
A closed-loop simulator for learning ultrasound acquisition renders a new B-mode frame at every agent step, so its texture stage must be fast and controllable. Realism-first diffusion is far too slow to runintheloop,andglobalsim-to-realtranslatorsuseasingleappearance map across the whole image, giving no per-tissue control. We present SonoSPADE, a real-time, CUDA-free per-tissue texture stage trained unpaired with no real labels, whose generator is conditioned on both a physicsrenderandthetissuelabelslice.Asegmenter,pretrainedforfree on aligned simulator pairs and then frozen, pseudo-labels the real pool foranOASIS-stylediscriminatoranddrivesaper-tissueconsistencyloss. A single pass runs at 15 to 85 frames per second with no CUDA, two tothreeordersofmagnitudefasterthandiffusion.OnthepublicKaggle abdominal set, relabeling a region changes only that region’s texture, while label-free methods do not react at all (label-swap locality ≈26, versus 0). Per-organ intensity matches real about 2× better than any learnedbaseline,andstructureispreserved2.4×better;onwhole-frame FID/KID it trails by design. In an in-the-loop proof of concept, on an organ-conditionedtaskscoredbyafixed real-trainedrecognizer,thepertissue reward gives the highest ground-truth liver visibility in all three seeds. Code is available at the project page3.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/SASHIMI_035.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: https://papers.miccai.org/miccai-2026-sat/supp/SASHIMI_035_supp.pdf
Link to Open Review
Open Review Page: https://openreview.net/forum?id=dx1vwmw4e9
BibTex
@InProceedings{IntTha_SonoSPADE_MICCAISAT2026,
author = { Intharah, Thanapong AND Gao, Zhichao AND Dong, Hao},
title = { { SonoSPADE: Real-Time, Per-Tissue Ultrasound Texture Synthesis for Closed-Loop Acquisition } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17258},
month = {pending},
page = {pending}
}