Super-Resolution on Rotationally Scanned Photoacoustic Microscopy Images Incorporating Scanning Prior

TL;DR

This work addresses the speed-accuracy trade-off in rotational photoacoustic microscopy (PAM) by proposing a super-resolution framework that leverages scanning priors. It combines a SURF-based registration to correct row displacements, a gradient-driven patch selection to emphasize vessel-rich regions, and a Transformer-based SR network (inspired by SwinIR) to achieve high-fidelity reconstructions. A scanning consistency loss is introduced to align edge structures between sampling-space SR outputs and image-space reconstructions, and vessel segmentation is used as a downstream evaluation to reflect clinical relevance. Experiments on synthetic and real undersampled PAM data demonstrate superior quantitative gains (PSNR/SSIM/MSE) and robust vessel-edge recovery, indicating strong potential for accelerating PAM imaging in practice. The framework is validated across multiple scale factors and includes ablations confirming the contributions of registration, patch selection, and scanning-prior loss, with public code provided for reproducibility.

Abstract

Photoacoustic Microscopy (PAM) images integrating the advantages of optical contrast and acoustic resolution have been widely used in brain studies. However, there exists a trade-off between scanning speed and image resolution. Compared with traditional raster scanning, rotational scanning provides good opportunities for fast PAM imaging by optimizing the scanning mechanism. Recently, there is a trend to incorporate deep learning into the scanning process to further increase the scanning speed.Yet, most such attempts are performed for raster scanning while those for rotational scanning are relatively rare. In this study, we propose a novel and well-performing super-resolution framework for rotational scanning-based PAM imaging. To eliminate adjacent rows' displacements due to subject motion or high-frequency scanning distortion,we introduce a registration module across odd and even rows in the preprocessing and incorporate displacement degradation in the training. Besides, gradient-based patch selection is proposed to increase the probability of blood vessel patches being selected for training. A Transformer-based network with a global receptive field is applied for better performance. Experimental results on both synthetic and real datasets demonstrate the effectiveness and generalizability of our proposed framework for rotationally scanned PAM images'super-resolution, both quantitatively and qualitatively. Code is available at https://github.com/11710615/PAMSR.git.

Figures (7)

• Figure 1: Inherent displacement degradation existed in PAM images. Both rotational scanning and raster scanning encounter displacements between adjacent rows.
• Figure 2: Our proposed framework for PAM image super-resolution. (a) illustrates the registration module. (b) depicts the architecture of our training model. (c) showcases the downstream segmentation task.
• Figure 3: Rotational scanning derived PAM images need to be reconstructed from the sampling space to the image space because of the zero-point issue.
• Figure 4: Visual comparisons on synthetic undersampled PAM images with scale factors of 2 (a), 4 (b) and 8 (c). Each sub-figure showcases the SR results in the first row, the corresponding segmentation results in the second row, and the error maps in the third row. Regions with obvious differences are highlighted by red rectangles.
• Figure 5: Visual comparisons of pixel intensity profiles along three small vessels: (a) to (c) show results from synthetic data with scale factors of 2, 4, and 8, respectively, while (d) displays results from real undersampled data. Our results demonstrate the closest resemblance to the ground truth.