Unsupervised Ultrafast Ultrasound Imaging Based on Decoupled Contrastive Learning

High-quality ultrafast ultrasound imaging typically relies on coherent compounding of multiple consecutive unfocused transmissions such as circular or plane waves (PWs), while hindering the gain in frame rate. Deep learning-based ultrasound imaging has emerged as a promising solution to improve ultrasound image quality. However, existing studies predominantly focus on supervised learning that relies on high-quality paired references. In this study, we introduce a high-quality ultrafast ultrasound imaging approach using unsupervised learning. The framework consists of one-sided image reconstruction based on adversarial learning for quality improvement, and decoupled contrastive learning for structural consistency. Experimental results on a public PW imaging dataset demonstrated that the proposed method yields high-quality images using single-angle PWs, competing with the coherent compounding of multi-angle PWs, which thereby indicates the feasibility of the proposed unsupervised learning framework for high-quality ultrafast ultrasound imaging.