UltraStar: Semantic-Aware Star Graph Modeling for Echocardiography Navigation

TL;DR

This paper proposes UltraStar, which reformulates probe navigation from path regression to anchor-based global localization by establishing a Star Graph, which treats historical keyframes as spatial anchors connected directly to the current view, explicitly modeling geometric constraints for precise positioning.

Abstract

Echocardiography is critical for diagnosing cardiovascular diseases, yet the shortage of skilled sonographers hinders timely patient care, due to high operational difficulties. Consequently, research on automated probe navigation has significant clinical potential. To achieve robust navigation, it is essential to leverage historical scanning information, mimicking how experts rely on past feedback to adjust subsequent maneuvers. Practical scanning data collected from sonographers typically consists of noisy trajectories inherently generated through trial-and-error exploration. However, existing methods typically model this history as a sequential chain, forcing models to overfit these noisy paths, leading to performance degradation on long sequences. In this paper, we propose UltraStar, which reformulates probe navigation from path regression to anchor-based global localization. By establishing a Star Graph, UltraStar treats historical keyframes as spatial anchors connected directly to the current view, explicitly modeling geometric constraints for precise positioning. We further enhance the Star Graph with a semantic-aware sampling strategy that actively selects the representative landmarks from massive history logs, reducing redundancy for accurate anchoring. Extensive experiments on a dataset with over 1.31 million samples demonstrate that UltraStar outperforms baselines and scales better with longer input lengths, revealing a more effective topology for history modeling under noisy exploration.

Figures (6)

• Figure 1: Comparison of modeling paradigms. (a) The echocardiography navigation task. (b) Single-frame methods struggle with localization due to limited context. (c) Sequential Graph methods model history as a chain, forcing the model to overfit noisy exploration trajectories and degrading localization accuracy. (d) Our Star Graph breaks the chain, treating historical keyframes as global anchors and learning direct geometric constraints for precise localization. (e) Our method achieves lower error and scales better.
• Figure 2: Illustration of the dataset. The view images are sourced from mitchell2019guidelines.
• Figure 3: Illustration of the Star Graph modeling paradigm.
• Figure 4: Diagram of segmental sampling and idea of semantic-aware sampling.
• Figure 5: Left: Scalability analysis. As the input graph size increases, our method demonstrates superior scalability with a consistent reduction in translation and rotation errors. Right: Ablation study on sampling strategies. We compare segmental sampling and semantic-aware sampling across different graph formulations, showing that semantic-aware sampling consistently yields lower translation and rotation errors.