Depth-aware transformer network for adaptive beamforming in single-angle plane-wave imaging

Single-angle plane-wave imaging (PWI) offers high imaging speed but suffers from low contrast and resolution, whereas multi-angle PWI improves image quality at the cost of slower acquisition. Recently, deep learning-based adaptive beamforming has been applied to enhance single-angle PWI; however, existing methods fail to fully exploit depth information, which critically affects aperture selection. To address this limitation, we propose a Depth-Aware Transformer Network (DATN), which leverages depth information and incorporates traditional apodization techniques. The DATN comprises two proposed modules: the Depth-Aware Gating (DAG) module and the Apodization Channel Self-Attention (ACSA) module. It operates in three steps: first, delayed data for each pixel are fed into the DAG module for feature extraction and depth-aware channel compression; second, the extracted features are processed by two cascaded ACSA modules with embedded Hanning window to estimate adaptive channel weights; finally, these weights are used to reconstruct the output image. Simulation, phantom, and in vivo experiments were conducted to evaluate DATN. In simulation, single-angle PWI reconstructed with DATN achieved a lateral full width at half maximum (FWHM) of 300 μm and a contrast-to-noise ratio (CNR) of 6.11. In phantom experiments, it achieved a lateral FWHM of 310 μm and a CNR of 3.17. In vivo carotid artery imaging shows that the proposed method outperforms DAS, CF, and MV, but remains inferior to 75-CPWC and slightly below 41-CPWC in terms of CNR; 75-CPWC is used as the training label for DATN. The proposed DATN markedly improves single-angle PWI quality by effectively integrating depth information and apodization techniques.