Abstract
Accurate segmentation of cardiac chambers in echocardiography videos is essential for quantitative functional assessment, yet remains challenging due to low image quality, rapid motion, and heavy annotation burden. While recent deep learning methods achieve high frame-wise accuracy, they often neglect temporal dependencies, resulting in temporally unstable predictions. Recent methods leverage temporal information or full 3D modeling to improve temporal coherency, but are often hindered by expensive computational costs or local temporal constraints. To address these limitations, we propose LaST-Diff, a latent diffusion–based framework for efficient and temporally stable echocardiography video segmentation. With a three-stage training strategy, LaST-Diff first learns strong spatial representations from images and then refines temporal coherence at the sequence level through a dedicated temporal branch with adaptive fusion, which enables effective spatiotemporal modeling without the computational burden. Experiments on both public and private datasets demonstrate that LaST-Diff achieves competitive static segmentation accuracy, while more faithfully modeling the anatomical trajectory across frames, leading to reduced inter-frame flickering and inconsistent boundary jitter. Notably, the robustness of LaST-Diff is particularly pronounced by the improvements on low-quality videos. Furthermore, the generative nature of LaST-Diff enables uncertainty quantification, offering critical confidence indicators for expert clinical review.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/DGM4MICCAI_028.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/profile?id=~Xiaojie_Chen7
BibTex
@InProceedings{CheXia_LaSTDiff_MICCAISAT2026,
author = { Chen, Xiaojie AND Zaman, Fahim Ahmed AND Chang, Amanda AND Dai, Liya AND Liu, Kan AND Sonka, Milan AND Wu, Xiaodong},
title = { { LaST-Diff: Latent Spatiotemporal Diffusion for Temporally Stable Echocardiography Video Segmentation } },
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}
}