Nanchang University has made progress in intelligent photoacoustic tomography imaging

Photoacoustic tomography (PAT) is a novel hybrid medical imaging technique that enables precise imaging of biological tissue structures at different spatial scales. It has been widely used in various fields, including brain imaging, cancer detection, and cardiovascular disease diagnosis. However, due to limitations in data acquisition conditions, photoacoustic tomography systems typically can only collect photoacoustic signals from a limited detection angle, which inevitably leads to a decrease in the image quality of photoacoustic tomography. How to achieve high-quality reconstruction under limited perspective sampling has always been an urgent problem that PAT needs to solve.

Recently, a research team from the Imaging and Visual Representation Laboratory at Nanchang University proposed a high-quality photoacoustic tomography imaging method based on a fractional diffusion model under limited viewing angles. The achievement was published in Photoacoustics, a top journal in the field of optoacoustics, under the title "Score based generative model assisted information compensation for high-quality limited view reconstruction in photoacoustic tomography".

Main research content
The research team proposed a photoacoustic tomography reconstruction method based on the fractional diffusion model. During the training phase, the model learns the data distribution of the samples by gradually adding noise to the existing samples. In the reconstruction stage, this method uses the prior information about image reconstruction learned by the diffusion model as the regularization term in the iterative reconstruction algorithm, and through cyclic iteration, high-quality photoacoustic tomography imaging under limited viewing angles can be achieved.

Figure 1. Process diagram of PAT reconstruction based on diffusion model method from a limited perspective.

As a validation, the research team evaluated the performance of the proposed method using experimental data from circular phantoms and live mice. In the circular phantom reconstruction experiment, this method was compared with traditional delay summation method (DAS), gradient descent method without regularization term (GD), gradient descent method with Tikhonov regularization term, U-Net method, and GAN method. The results are shown in Figure 2. The proposed method shows higher quality and clearer contours in the reconstruction results under different limited viewing angles. At a limited viewing angle of 70 °, the proposed method achieved peak signal-to-noise ratio (PSNR) and structural similarity (SSIM) of 31.57dB and 0.95, respectively, which were improved by 203% and 48% compared to the delay summation method.

Figure 2. Reconstruction results of circular phantom.
From the experimental results of the simulated small balls and live mice (experimental data), it can be seen that this method still has good performance (Figure 3). Specifically, in extremely limited detection angles (such as a 90 ° limited angle), this method outperforms the U-Net method significantly. In live mouse experiments, this method achieved an SSIM/PSNR of 0.80/29.18 dB in reconstructed images with a limited viewing angle of 90 °. Compared to the U-Net method, the PSNR increased by 64% and the SSIM increased by 48%.

Figure 3. Reconstruction results of live data from different detection perspectives.

Conclusion and Prospect
This study proposes a new high-quality photoacoustic tomography imaging strategy based on the fractional diffusion model under limited viewing angles. This method combines the physical model of PAT with the diffusion model, and introduces high-dimensional prior information learned by the diffusion model deep network in the model-based iteration process. This method significantly improves the imaging quality and effectively solves the problem of image quality degradation caused by limited viewing angle sampling in PAT, with the potential to accelerate PAT imaging speed and expand its application range.

Guo Kangjun, master's student Zheng Zhiyuan, master's student Zhong Wenhua, and master's student Li Zilong from Nanchang University are co first authors of the article. Professor Liu Qiegen and Associate Professor Song Xianlin are co corresponding authors. This study was supported by the National Natural Science Foundation of China and the Key Research and Development Project of Jiangxi Province.

Source: Opticsky