SZU PolyU
CVPR 2026

OSA: Echocardiography Video Segmentation via
Orthogonalized State Update and Anatomical Prior-aware Feature Enhancement

Rui Wang1,  Huisi Wu1*,  Jing Qin2
1College of Computer Science and Software Engineering, Shenzhen University
2Centre for Smart Health, School of Nursing, The Hong Kong Polytechnic University
SZU PolyU

Contents

01
Problem Setting
02
Method
03
Experiments
04
Discussion
2
SZU PolyU

Problem Setting

Medical

(a)Speckle Noise, (b)Blurred Contours, and (c-f)Pronounced Variations in the Target's Morphology Across the Cardiac Cycle

Video

(a)Extended Temporal Contexts, (b)Efficiency–Accuracy Trade-off in Recall, (c)Computational Burden

Task

Clinical Simulation Setting — Absence of Ground Truth at Inference; Training via Boundary-Frame Prediction and Loss Computation

Echocardiography frames showing speckle noise, blurred contours, and morphological variations across the cardiac cycle Visual illustration of challenges in echocardiography video segmentation
3
SZU PolyU

Method — Overview

OSA method overview
Fig.1: Overall pipeline of OSA — Orthogonalized State Update + Anatomical Prior-aware Feature Enhancement
4
SZU PolyU

Method — Orthogonalized State Update

StEuc=St−1(αt(ICk−βtktkt⊤))+βtvtkt⊤.\mathbf{S}_t^{\text{Euc}} = \mathbf{S}_{t-1}\left(\alpha_t (\mathbf{I}_{C_k} - \beta_t \mathbf{k}_t \mathbf{k}_t^\top)\right) + \beta_t \mathbf{v}_t \mathbf{k}_t^\top.
VCv,Ck={S∈RCv×Ck:S⊤S=ICk}.\mathcal{V}_{C_v, C_k} = \{\mathbf{S} \in \mathbb{R}^{C_v \times C_k} : \mathbf{S}^\top \mathbf{S} = \mathbf{I}_{C_k}\}.
St=Proj⁡V(StEuc)=arg⁡min⁡S∈VCv,Ck12∥S−StEuc∥F2.\mathbf{S}_t = \operatorname{Proj}_{\mathcal{V}}(\mathbf{S}_t^{\text{Euc}}) = \underset{\mathbf{S} \in \mathcal{V}_{C_v, C_k}}{\arg \min} \frac{1}{2} \|\mathbf{S} - \mathbf{S}_t^{\text{Euc}}\|_F^2.
X(0)=StEuc∥StEuc∥F+ϵ.\mathbf{X}^{(0)} = \frac{\mathbf{S}_t^{\text{Euc}}}{\|\mathbf{S}_t^{\text{Euc}}\|_F + \epsilon}.
Stiefel manifold optimization illustration
5
SZU PolyU

Method — Anatomical Prior-aware Feature Enhancement

Mt=AvgPoolK×K(Xt)\mathbf{M}_t = \mathrm{AvgPool}_{K \times K}(\mathbf{X}_t)
Xt+=ReLU(Xt−Mt),Xt−=ReLU(Mt−Xt)\mathbf{X}_t^{+} = \mathrm{ReLU}(\mathbf{X}_t - \mathbf{M}_t), \quad \mathbf{X}_t^{-} = \mathrm{ReLU}(\mathbf{M}_t - \mathbf{X}_t)
Ht+=ϕ+(Xt+),Ht−=ϕ−(Xt−)\mathbf{H}_t^{+} = \phi^{+}(\mathbf{X}_t^{+}), \quad \mathbf{H}_t^{-} = \phi^{-}(\mathbf{X}_t^{-})
λt=σ ⁣(Wg[Ht+; Ht−]),Zt=λt⊙Ht++(1−λt)⊙Ht−\begin{aligned}\lambda_t &= \sigma\!\left( \mathbf{W}_g \left[\mathbf{H}_t^{+};\, \mathbf{H}_t^{-}\right] \right), \\ \mathbf{Z}_t &= \lambda_t \odot \mathbf{H}_t^{+} + (1 - \lambda_t) \odot \mathbf{H}_t^{-}\end{aligned}
Anatomical Prior-aware Feature Enhancement diagram
6
SZU PolyU

Experiments

Quantitative comparison of OSA against baseline methods on CAMUS and EchoNet-Dynamic datasets
01

OSA achieves state-of-the-art Dice on CAMUS & EchoNet-Dynamic.

02

OSU stabilizes hidden-state trajectories on Stiefel manifold.

03

APFE enhances feature discrimination via anatomical priors.

04

Lightweight design: only +8.8% params vs. GDKVM (38.3 vs 35.2 M).

7
SZU PolyU

Experiments

Ablation study results for OSU and APFE modules Quantitative ablation results table comparing module contributions
8
SZU PolyU

Discussion

•

Generality: Extend to broader ultrasound datasets.

•

Longer sequences: Tackle longer videos, complex rhythms, difficult cases.

•

Hardware: Optimize matrix state for parallel acceleration.

9
SZU PolyU

Conclusion

1

Propose OSA: orthogonalized state updates for echocardiography video segmentation.

2

OSU prevents rank collapse via Stiefel manifold constraint.

3

APFE decouples anatomical structures from speckle noise.

4

State-of-the-art on CAMUS and EchoNet-Dynamic.

10
SZU PolyU

Limitation

•

Fixed-length: Constrained to fixed-length sequences.

•

Online inference: Stable memory propagation is difficult.

•

Failure cases: May occur in some situations.

•

Domain shift: May lack robustness to varying equipment.

Failure cases on CAMUS and EchoNet-Dynamic
11
SZU PolyU
CVPR 2026

Thanks!

12