USRMamba: Adaptive Routing-Guided State Space Model for Ultrasound Super-Resolution

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Signal Processing

Background:

• Ultrasound (US) imaging resolution is limited by acoustic diffraction and transducer density, impacting clinical diagnosis.
• Super-resolution (SR) reconstruction offers a cost-effective alternative to system upgrades for improving US image quality.
• Existing SR methods struggle with the complex acoustic properties of tissues, hindering unified model development.

Purpose of the Study:

• To pioneer a novel Mamba-based single US image SR method, named USRMamba.
• To enhance the fidelity and diagnostic utility of ultrasound images through advanced SR reconstruction.
• To overcome limitations in current SR techniques for ultrasound imaging.

Main Methods:

• Developed USRMamba, a Mamba-based SR method for single ultrasound images.
• Introduced an Enhanced Transform Combine Module (ETCM) for multi-scale feature extraction, addressing high-frequency loss and speckle noise.
• Proposed an Adaptive Top-k Prompt Module (ATPM) using adaptive routing to mitigate fuzzy region interference caused by attenuation.
• Integrated a Frequency Channel Attention Module (FCAM) for parallel frequency-spatial domain reconstruction.

Main Results:

• USRMamba demonstrated superior performance on various US datasets compared to existing methods.
• The method achieved an average PSNR of 1.31dB higher than state-of-the-art (SOTA) methods at a ×2 scale factor.
• Qualitative and quantitative experiments validated the effectiveness of USRMamba in optimizing US image SR reconstruction.

Conclusions:

• USRMamba represents a revolutionary approach to single US image SR reconstruction.
• The proposed method effectively enhances image quality and detail reconstruction in ultrasound imaging.
• USRMamba shows significant potential for improving clinical diagnosis through superior ultrasound image resolution.