Contour-Guided Query-Based Feature Fusion for Boundary-Aware and Generalizable Cardiac Ultrasound Segmentation

2026-03-30

AI Summary

This work addresses the challenges of ambiguous segmentation boundaries and structural inconsistency in cardiac ultrasound images, which arise from low contrast, speckle noise, and cross-domain shifts. To tackle these issues, the authors propose CGQR-Net, a novel architecture that leverages HRNet as its backbone to preserve fine spatial details. The method innovatively encodes anatomical contours as learnable query embeddings and employs a cross-attention mechanism to enable structure-aware fusion between these queries and multi-resolution features, thereby guiding precise boundary refinement. Coupled with a dual-head supervision strategy, the model achieves significant improvements in segmentation accuracy, boundary fidelity, and cross-dataset generalization on both the CAMUS and CardiacNet benchmarks.

Abstract

Accurate cardiac ultrasound segmentation is essential for reliable assessment of ventricular function in intelligent healthcare systems. However, echocardiographic images are challenging due to low contrast, speckle noise, irregular boundaries, and domain shifts across devices and patient populations. Existing methods, largely based on appearance-driven learning, often fail to preserve boundary precision and structural consistency under these conditions. To address these issues, we propose a Contour-Guided Query Refinement Network (CGQR-Net) for boundary-aware cardiac ultrasound segmentation. The framework integrates multi-resolution feature representations with contour-derived structural priors. An HRNet backbone preserves high-resolution spatial details while capturing multi-scale context. A coarse segmentation is first generated, from which anatomical contours are extracted and encoded into learnable query embeddings. These contour-guided queries interact with fused feature maps via cross-attention, enabling structure-aware refinement that improves boundary delineation and reduces noise artifacts. A dual-head supervision strategy jointly optimizes segmentation and boundary prediction to enforce structural consistency. The proposed method is evaluated on the CAMUS dataset and further validated on the CardiacNet dataset to assess cross-dataset generalization. Experimental results demonstrate improved segmentation accuracy, enhanced boundary precision, and robust performance across varying imaging conditions. These results highlight the effectiveness of integrating contour-level structural information with feature-level representations for reliable cardiac ultrasound segmentation.

Problem

• cardiac ultrasound segmentation
• boundary precision
• domain shift
• speckle noise
• structural consistency

Innovation

• contour-guided query
• cross-attention refinement
• boundary-aware segmentation
• structural prior
• multi-resolution fusion