AI 中文总结
本文提出角度域零值相减成像（角度 NSI），其基于平面波复合的角度图像操作，可缩窄横向宽度，计算时间接近延迟求和（DAS）且快于传统 NSI，兼容多数波束形成器。
AI 中文摘要
零值相减成像（Null subtraction imaging, NSI）通过结合零均值 apodization（变迹）及其两个直流偏移变体产生的声场，来缩窄表观横向响应。传统 NSI 将这些权重应用于接收阵元，因此需要获取通道数据和灵活的波束形成器，同时会增加计算时间。本文研究一种角度域类似方法，称为角度 NSI，其替代方案是在相干平面波复合过程中生成的每个角度图像上进行操作。采用矩阵公式区分接收域和角度域零值相减表达式，并识别它们的独立计算组件。该方法通过模拟点目标、开放微泡（microbubble, MB）轨迹数据集以及两次体内颈动脉采集数据进行评估。模拟中对角度采样、噪声、相位抖动和目标位置进行了扫描。同步 GPU 基准测试比较了完整重建结果，以传统接收域 NSI 和标准延迟求和（delay-and-sum, DAS）成像为参考。角度域 NSI 将模拟的 -6dB 横向宽度从 DAS 的 0.226 mm 缩小至 0.116 mm，而传统接收域 NSI 为 0.009 mm，所有方法的轴向宽度均为 0.721 mm。在 RTX A2000 GPU 上的中位数重建时间，DAS 为 23.92 ms，传统 NSI 为 30.19 ms，角度域 NSI 为 24.74 ms。在功率多普勒成像中，与 DAS 相比，传统 NSI 使 MB 轨迹的表观宽度缩小 50.2%，角度 NSI 缩小 30.9%。体内实验中，两种 NSI 变体均降低了相对于 DAS 的广义对比度噪声比。角度域 NSI 可作为所有平面波序列在波束形成后的后处理步骤简单实现，该技术与大多数常规波束形成器实现兼容，能有效缩窄表观横向宽度，且计算时间接近 DAS，比传统 NSI 更快。
英文摘要
Null subtraction imaging (NSI) narrows the apparent lateral response by combining fields produced with a zero-mean apodization and two DC-offset variants. Conventional NSI applies these weights across receive elements and therefore requires access to channel data and flexible beamformers, while increasing the computation time. We investigate an angle-domain analogue, termed angular NSI, that instead operates on the per-angle images produced during coherent plane-wave compounding. A matrix formulation was used to distinguish receive-domain and angle-domain null subtraction expressions and to identify their independent computational components. The method was evaluated using a simulated point target, an open microbubble (MB) traces dataset and two in vivo carotid acquisitions. Angle sampling, noise, phase jitter and target position were swept in simulation. A synchronized GPU benchmark compared the complete reconstructions. Conventional receive-domain NSI and standard delay-and-sum (DAS) imaging were references. Angle-domain NSI reduced the simulated -6dB lateral width from 0.226 mm for DAS to 0.116 mm, compared to 0.009mm for conventional receive-domain NSI, while the axial width was 0.721 mm for all methods. Median reconstruction times on an RTX A2000 GPU were 23.92, 30.19 and 24.74 ms for DAS, conventional NSI and angle-domain NSI. In power Doppler, the apparent width of the MB traces were reduced by 50.2% for conventional NSI and 30.9% for angular NSI compared to DAS. In vivo, both NSI variants reduced generalized contrast-to-noise ratio relative to DAS. Angle-domain NSI can be simply implemented as a post-processing step for all plane-wave sequences after beamforming. The technique is compatible with most of the usual beamformer implementations. It effectively reduces apparent lateral width, and its computation time is close to DAS and faster than conventional NSI.
