Deep unfolding networks (DUNs) are the foremost methods in the realm of compressed sensing MRI, as they can employ learnable networks to facilitate interpretable forward-inference operators. However, several daunting issues still exist, including the heavy dependency on the first-order optimization algorithms, the insufficient information fusion mechanisms, and the limitation of capturing long-range relationships. To address the issues, we propose a Generically Accelerated Half-Quadratic Splitting (GA-HQS) algorithm that incorporates second-order gradient information and pyramid attention modules for the delicate fusion of inputs at the pixel level. Moreover, a multi-scale split transformer is also designed to enhance the global feature representation. Comprehensive experiments demonstrate that our method surpasses previous ones on single-coil MRI acceleration tasks.

翻译：深度展开网络（DUNs）是压缩感知磁共振成像领域最前沿的方法，其利用可学习网络构建可解释的前向推理算子。然而，当前仍存在若干棘手问题，包括对一阶优化算法的严重依赖、信息融合机制不充分以及长程关系捕获能力受限。针对这些问题，我们提出一种通用加速半二次分裂（GA-HQS）算法，该算法融合二阶梯度信息与金字塔注意力模块，实现像素级输入的精细融合。此外，还设计了多尺度分裂Transformer以增强全局特征表征能力。综合实验表明，本方法在单线圈磁共振加速任务中超越了以往的所有方法。