基于深度学习理论的光学分子成像图像重建方法  

Image reconstruction of optical molecular imaging based on deep learning theory

作　　者：周媛媛[1] 王杨丽[2] ZHOU Yuanyuan;WANG Yangli(Foreign trade College of Chongqing Normal University,Information Technology Center,Chongqing 401520,China;Foreign trade College of Chongqing Normal University,College of Media Convergence,Chongqing 401520,China)

机构地区：[1]重庆师范大学涉外商贸学院信息技术中心,重庆401520 [2]重庆师范大学涉外商贸学院影视融媒体学院,重庆401520

出　　处：《激光杂志》2021年第12期114-118,共5页Laser Journal

基　　金：重庆市教委高等教育教改项目(No.203516)。

摘　　要：光学分子成像图像重建是当前的研究重点,由于传统光学分子成像图像重建方法存在重建误差,效果不理想等缺陷,为了获得理想的光学分子成像图像重建效果,提出了基于深度学习理论的光学分子成像图像重建方法。首先分析光学分子成像图像重建工作原理,找到导致光学分子成像图像重建质量差的因素,然后采集光学分子成像图像,选择卷积神经网络算法进行光学分子成像的图像重建。实验结果显示:采用本方法进行峰值信噪比(PSNR)范围为34.16~38.96,结构相似性(SSIM)范围为0.8549~0.9808,均远远大于常规方法数值,表明提出方法图像重建质量较高,充分证明提出方法性能较好。应用价值相似性指标更大,图像重建质量更高,则提出方法的图像重建性能更佳。Optical molecular imaging image reconstruction is the focus of current research.Due to the defects of traditional optical molecular imaging image reconstruction methods,such as reconstruction error and unsatisfactory effect,in order to obtain reasonable reconstruction effect of optical molecular imaging image,an optical molecular imaging image reconstruction method based on deep learning theory was proposed.Firstly,the working principle of image reconstruction of optical molecular imaging is analyzed,and the factors leading to poor quality of image reconstruction of optical molecular imaging are found.Then,the optical molecular imaging images are collected and the convolutional neural network algorithm is selected to reconstruct the image of optical molecular imaging.Experimental results show that the range of peak signal-to-noise ratio(PSNR)and structural similarity(SSIM)of the proposed method is 34.16-38.96 and 0.8549-0.9808,both of which are far greater than the values of conventional methods,indicating that the proposed method has a high quality of image reconstruction and fully proves that the proposed method has a good performance.If the similarity index of applied value is larger and the quality of image reconstruction is higher,the image reconstruction performance of the proposed method is better.

关 键 词：光学技术 分子成像图像 卷积神经网络 图像重建 

分 类 号：TN929[电子电信—通信与信息系统]