Zhang Yu, Li Shuang, Wang Yibing, Sun Yu, Huang Tingting, Xiang Wenyi, Li Changhui
Department of Biomedical Engineering, College of Future Technology, Peking University, Beijing, China.
School of Electrical and Electronic Engineering, Nanyang Technological University, , Singapore, Singapore.
Photoacoustics. 2025 Apr 26;44:100726. doi: 10.1016/j.pacs.2025.100726. eCollection 2025 Aug.
In reality, photoacoustic imaging (PAI) is generally influenced by artifacts caused by sparse array or limited view. In this work, to balance the computing cost and artifact removal performance, we propose an iterative optimization method that does not need to repeat solving forward model for every iteration circle, called the regularized iteration method with structural prior (RISP). The structural prior is a probability matrix derived from multiple reconstructed images via randomly selecting partial array elements. High-probability values indicate high coherency among multiple reconstruction results at those positions, suggesting a high likelihood of representing true imaging results. In contrast, low-probability values indicate higher randomness, leaning more towards artifacts or noise. As a structural prior, this probability matrix, together with the original PAI result using all array elements, guides the regularized iteration of the PAI results. The simulation and real animal and human PAI study results demonstrated our method can substantially reduce image artifacts, as well as noise.
实际上，光声成像（PAI）通常会受到稀疏阵列或有限视角所导致的伪影影响。在这项工作中，为了平衡计算成本和伪影去除性能，我们提出了一种迭代优化方法，该方法无需在每个迭代循环中重复求解正向模型，称为具有结构先验的正则化迭代方法（RISP）。结构先验是一个概率矩阵，它通过随机选择部分阵列元素从多个重建图像中推导得出。高概率值表明在这些位置的多个重建结果之间具有高相干性，这表明更有可能代表真实的成像结果。相反，低概率值表明更高的随机性，更倾向于伪影或噪声。作为一种结构先验，这个概率矩阵与使用所有阵列元素的原始PAI结果一起，指导PAI结果的正则化迭代。模拟以及真实动物和人体PAI研究结果表明，我们的方法可以大幅减少图像伪影以及噪声。
