Abstract
Background/Objectives: Single plane wave imaging (SPWI) offers ultrafast acquisition rates suitable for real-time ultrasound imaging applications; however, its image quality is compromised by beamforming artifacts such as sidelobe and grating lobe interferences. Methods: In this paper, we introduce an unsupervised beamforming framework based on adaptive deep coherence loss (DCL-A), which employs linear (αlinear) or nonlinear weighting (αnonlinear) within the coherence loss function to enhance the artifact suppression and improve overall image quality. During training, the adaptive weight (α) is determined by the angular distance between the input and target PW frames, assigning lower α values for smaller distances and higher α values for larger distances. Therefore, this adaptability enables the method to surpass conventional DCL (no weighting) by emphasizing the different spatial correlation characteristics of mainlobe and sidelobe signals. To assess the performance of the proposed method, we trained and validated the network using publicly available datasets, including simulation, phantom and in vivo images. Results: In the simulation and phantom studies, the DCL-A with αnonlinear outperformed the comparison methods (i.e., conventional DCL and DCL-A with αlinear) in terms of peak range sidelobe level (PRSLL), achieving 7 dB and 14 dB greater sidelobe suppression, respectively, while maintaining a comparable full width at half maximum (FWHM). In the in vivo study, it achieved the highest contrast resolution among the comparison methods, yielding 2% and 3% improvements in generalized contrast-to-noise ratio (gCNR), respectively. Conclusions: These results demonstrate that the proposed deep learning-based beamforming framework can significantly enhance SPWI image quality without compromising frame rate, indicating promising potential for high-speed, high-resolution clinical applications such as cardiac assessment and real-time interventional guidance.