Hong Xiaobin, Tang Furong, Wang Lidai, Chen Jiangbo
School of Mechanical & Automotive Engineering, South China University of Technology, Guangzhou, Guangdong, PR China.
Department of Biomedical Engineering, City University of Hong Kong, 83 Tat Chee Ave, Kowloon, Hong Kong Special Administrative Region of China.
Photoacoustics. 2024 Jul 5;38:100632. doi: 10.1016/j.pacs.2024.100632. eCollection 2024 Aug.
A fast scanner of optical-resolution photoacoustic microscopy is inherently vulnerable to perturbation, leading to severe image distortion and significant misalignment among multiple 2D or 3D images. Restoration and registration of these images is critical for accurately quantifying dynamic information in long-term imaging. However, traditional registration algorithms face a great challenge in computational throughput. Here, we develop an unsupervised deep learning based registration network to achieve real-time image restoration and registration. This method can correct artifacts from B-scan distortion and remove misalignment among adjacent and repetitive images in real time. Compared with conventional intensity based registration algorithms, the throughput of the developed algorithm is improved by 50 times. After training, the new deep learning method performs better than conventional feature based image registration algorithms. The results show that the proposed method can accurately restore and register the images of fast-scanning photoacoustic microscopy in real time, offering a powerful tool to extract dynamic vascular structural and functional information.
光学分辨率光声显微镜的快速扫描仪本质上容易受到干扰，从而导致严重的图像失真以及多个二维或三维图像之间出现明显的错位。这些图像的恢复和配准对于在长期成像中准确量化动态信息至关重要。然而，传统的配准算法在计算吞吐量方面面临巨大挑战。在此，我们开发了一种基于无监督深度学习的配准网络，以实现实时图像恢复和配准。该方法可以校正B扫描失真产生的伪影，并实时消除相邻和重复图像之间的错位。与传统的基于强度的配准算法相比，所开发算法的吞吐量提高了50倍。经过训练后，新的深度学习方法比传统的基于特征的图像配准算法表现更好。结果表明，所提出的方法能够实时准确地恢复和配准快速扫描光声显微镜的图像，为提取动态血管结构和功能信息提供了一个强大的工具。
