基于无监督学习的高分辨光声重建

High resolution optoacoustic reconstruction based on unsupervised learning

光声断层成像(Optoacoustic Tomography,OAT)是一种新兴的生物医学成像技术,在基础医学研究与临床实践中具有重要作用。针对现有光声断层成像空间分辨率较低的问题,提出了一种结合物理点扩散函数(Point Spread Func-tion,PSF)模型和卷积神经网络(Convolutional Neural Network,CNN)的新型高分辨光声重建网络方法(Physical Atten-tion U-Net,Phys-AU-Net)。该方法采用无监督学习策略,结合物理PSF模型和基于注意力机制的U-Net网络。其中,物理PSF模型用于完成对衍射受限机制的模拟,基于注意力机制的U-Net网络用于实现对高密度重叠吸收体图像的特征提取。在二者共同作用下,Phys-AU-Net突破了声衍射极限对于OAT成像空间分辨率的限制。实验结果表明,Phys-AU-Net能够有效实现对声衍射受限光声断层图像的高分辨重建,其性能相较于U-Net网络具有较大程度提升,在结构相似性指标(Structural Similarity,SSIM)方面提升了43.5%,在峰值信噪比(Peak Signal to Noise Ratio,PSNR)方面提升了11.2%,对临床研究及诊断具有重要意义。

Optoacoustic tomography(OAT)is a new biomedical imaging technology,which plays an important role in medical research and clinical practice.Considering the problems of low resolution in optoacoustic tomography,a new high-resolution reconstruction network(Phys-AU-Net)combining physical point spread function(PSF)and convolutional neural network(CNN)is proposed in this paper.Briefly,the proposed method adopts an unsupervised learning strategy,and combines a point spread function(PSF)model and the U-Net based on the attention mechanism.Among these,the PSF model is used to simulate the diffraction limited mechanism,and the U-Net network based on attention mechanism is used to complete the feature extraction of high-density images.The experimental results show that compared with U-Net,the Phys-AU-Net improves the structural similarity(SSIM)by 43.5%and the peak signal to noise ratio(PSNR)by 11.2%,which provides a great potential in clinical research and diagnosis.