基于卷积神经网络和预处理交替投影的单光子发射计算机断层扫描图像重建

Single Photon Emission Computed Tomography Reconstruction Based on Convolutional Neural Network and Pre-Processing Alternating Projection

摘要: 单光子发射计算机断层扫描在肿瘤成像中发挥着重要作用，其图像重建是一个重要研究方向。预处理交替投影算法在单光子发射计算机断层扫描图像重建方面有良好的性能，但随着卷积神经网络在图像处理领域的发展以及业内对重建图像质量的追求逐年提升，预处理交替投影的重建效果已经较难满足业内的期望。改进成像硬件设备可以提升图像质量但是成本高昂且时间周期长。因此本文使用人体模型数据集，通过改进的卷积神经网络学习预处理交替投影重建图像与该数据集重建图像真图标签的映射，提升了重建图像的质量。实验结果表明，使用本方法在该数据集中的单光子发射计算机断层扫描图像重建，PSNR、MSE、SSIM数值指标均领先于预处理交替投影和滤波反投影法，同时在视觉效果、噪声抑制方面均超越上述对比方法。

Abstract: Single photon emission computed tomography plays an essential role in cancer imaging. Thereinto, image reconstruction is an important research direction. Preprocessing alternating projection algorithm applies well to single photon emission computed tomography image reconstruction. However, with the development of the convolutional neural network in the field of image processing and the pursuit of reconstructed image quality in the industry year by year, the reconstruction effect of preprocessing alternate projection has been challenging to meet the expectation of the industry. Improving the imaging hardware can improve the image quality, but the cost is high, and the time cycle is long. Therefore, this paper uses a mannequin dataset; an improved convolutional neural network is used to learn the mapping between the reconstructed image of alternating projection and the ground truth image label of the reconstructed image of the dataset, improving the image quality image reconstruction. The experimental results show that using this method in the image reconstruction of this data set, the numerical indicators PSNR, MSE and SSIM are superior to preprocessing alternating projection and filtered back projection. Meanwhile, this method is supe-rior to the above methods in visual effects and noise suppression.