Photoacoustic imaging is an emerging medical imaging technique, with photoacoustic microscopy (PAM) being based on the photoacoustic effect to observe optically absorbing structures. It achieves high lateral resolution through fine optical focusing and has shown significant potential in the study of the microstructure and function of biological tissues. However, PAM faces the challenge of slow imaging speeds when acquiring large-scale, high-resolution images. To address this, we develop a PAM system utilizing hybrid optical-mechanical scanning technology, enabling the automated acquisition of multiple image blocks at different positions to achieve large-scale image acquisition. Building on this, we further develop a stitching method specifically tailored for photoacoustic microvascular images, capable of overcoming challenges posed by vascular complexity and uneven signal detection. OR-PAM not only captures vascular signals but also various other signals, such as pigment signals, which often exhibit a mottled and complex structure with high similarity. This structural complexity increases the risk of mismatches during image stitching. Therefore, the development of a stable, efficient, and high-performance image stitching algorithm designed specifically for photoacoustic microscopy is crucial for achieving high-resolution, large-scale vascular imaging. Such an algorithm is a core technical advancement necessary to drive the development and application of photoacoustic microscopy technology.

In this paper, we develop a high-resolution, high-speed PAM system based on hybrid optical-mechanical scanning and propose a stitching method tailored for vascular images in photoacoustic microscopy. Using this system, we acquire several sub-image blocks of carbon fiber phantoms and in vivo samples. By applying our proposed stitching method, we process these sub-image blocks, demonstrating the feasibility of our approach even under conditions where the overlap rate is as low as 3% for the carbon fiber phantoms.

We present imaging results of a carbon fiber phantom with a diameter of approximately 6 μm. A total of 54 sub-image blocks are acquired, and after stitching, the resulting image has a size of approximately 3.5 mm×2.3 mm, with an overlap rate of only 3% between adjacent sub-image blocks [Figs. 4(a) and (b)]. The total acquisition time, including data collection, transfer, and platform movement, is approximately 72 s. The stitching is precise, with no noticeable stripe artifacts. The brain imaging results involve the acquisition of 81 sub-image blocks [Fig. 5(a)]. The fused image has a size of approximately 3.2 mm×3.2 mm, with an overlap rate of 10% between adjacent blocks. The stitched image exhibits no noticeable stripe artifacts, with continuous vasculature and visible capillaries [Fig. 5(b)]. Finally, the original sub-image block, the sub-image block processed by median filtering and the NLM algorithm based on spatial denoising, and the final processed sub-image block are shown (Fig. 6). The signal-to-noise ratio (SNR) of the image improves from 9 dB to 29.3 dB, with a final SNR of 35.9 dB. The noise is greatly suppressed, significantly enhancing the visual quality of the image.

In summary, we present a fast PAM system based on hybrid optical-mechanical scanning and a stitching method suitable for photoacoustic images. The system’s imaging and image stitching capabilities are validated through phantom and in vivo experiments. This system is significant for improving imaging speed and enlarging the field of view. Notably, it employs a miniature transducer to receive ultrasound signals, offering the advantages of compactness, high sensitivity, and rapid response. The system can be extended to an array of transducers to further increase the field of view. In addition, the 2D galvanometer can be placed in air, avoiding the need for the scanning components to be submerged in water as required in existing high-speed PAM systems. This feature reduces the influence of water disturbances and offers improved stability. In terms of image processing, the spatial denoising NLM has shown great potential in PAM image denoising, effectively removing spatial noise. BaSiC, one of the most advanced shadow correction algorithms, performs exceptionally well on PAM images. The phase correlation-based stitching algorithm can avoid the issues of sparse feature points and mismatches typically encountered in feature-based stitching algorithms when the overlap rate is extremely low. Since the motorized stage moves the same distance along a set direction each time and the field of view is fixed, no additional image registration is required. In the field of PAM, we present a practical method to expand the field of view and increase imaging speed.