Nanchang University research progresses in acoustic resolution photoacoustic microimaging enhancement

As a promising imaging modality that combines the high spatial resolution of optical imaging and the deep tissue penetration ability of ultrasound imaging, photoacoustic microscopy (PAM) has attracted a lot of attention in the field of biomedical research, and has a wide range of applications in many fields, such as tumor detection, dermatology, and vascular morphology assessment. Depending on the imaging modality, PAM can be categorized into optical resolution photoacoustic microscopy (OR-PAM) and acoustic resolution photoacoustic microscopy (AR-PAM.) OR-PAM utilizes optically strong focusing to achieve high lateral resolution (<5 μm) imaging. However, light scattering within biological tissues limits the penetration depth of OR-PAM (no more than 1-2 mm). In contrast, AR-PAM exhibited deeper imaging (~3-10 mm). However, this enhanced effect is accompanied by a decrease in lateral resolution (>50 μm) and an increase in background noise. How to realize AR-PAM imaging with high lateral resolution without sacrificing imaging depth has been a pressing problem for PAM.

Recently, a research team from the Laboratory of Imaging and Visual Representation, Nanchang University, proposed an acoustic-resolution photoacoustic microscopy enhancement strategy based on the mean-reverting diffusion model to realize the transition from acoustic resolution to optical resolution. The result is published as “Mean-reverting diffusion model-enhanced acoustic-resolution photoacoustic microscopy for resolution enhancement: toward The results were published in the Journal of Innovative Optical Health Sciences, a leading journal in the field of biomedical photonics, under the title of “optical resolution”.

Main research content

The research team proposes a mean-regression diffusion model-based enhancement strategy for acoustic-resolution photoacoustic microscopy to achieve enhancement from acoustic to optical resolution. In the training stage, a mean-reversion diffusion model is trained to learn a priori information about the data distribution by modeling the quality reduction process from high-resolution PAM images to low-resolution AR-PAM images with fixed Gaussian noise. In the reconstruction stage, the learned a priori information is used to iteratively sample the noise states to generate a high resolution image from the low quality AR-PAM image. 

Figure 1.Flowchart of AR-PAM enhancement algorithm based on mean-reversion diffusion modeling

As a validation, the research team evaluated the performance of the proposed method using in vivo mouse experimental data. In a scene with a lateral resolution of 55 μm and a signal-to-noise ratio (SNR) of 35 dB, the method was compared with the conventional RL deconvolution method, the CycleGAN method, and the FDUnet method, and the results are shown in Figure 2. The enhancement results of the proposed method show higher quality and superior lateral resolution. The Peak Signal to Noise Ratio (PSNR) and Structural Similarity (SSIM) reached 31.96 dB, 0.91, respectively, which is 136% and 54% improvement over the RL deconvolution method. 

Fig. 2. Comparison of the reconstruction results of different methods.

In addition, in order to further validate the enhancement performance of the model on large-scale images, the research team also used complete live mouse cerebrovascular images for experiments. It can be seen that the enhanced image (as shown in Fig. 3(c)) has a clearer vascular topology, higher lateral resolution, and stronger image contrast. It is noteworthy that the continuity of the vessels at the sub-image joints is well maintained without obvious artifacts. Compared with the true-value image (as shown in Fig. 3(a)), the PSNR and SSIM of the AR-PAM image (as shown in Fig. 3(b)) were 19.39 dB and 0.53, respectively, and those of the model-enhanced image were improved to 24.72 dB and 0.73, respectively, which were 27% and 38% higher compared to the AR-PAM, respectively. The results show that the proposed method can still significantly improve the lateral resolution of large-size AR-PAM images. 

Fig.3. Resolution enhancement results for large size AR-PAM images.

Conclusion and Outlook

This study proposes a new AR-PAM enhancement strategy based on a mean-reversion diffusion model to achieve a balance between AR-PAM and OR-PAM imaging depth and lateral resolution. The method models the quality reduction process from OR-PAM to low quality AR-PAM images. Subsequently, a numerical method is used to iteratively perform an inverse-time SDE aimed at reconstructing high-quality images from a homogenized state. The method significantly improves the lateral resolution of AR-PAM without sacrificing the imaging depth, which has the potential to improve the quality of PAM imaging and extend its application range.

Source: opticsky