一种细胞荧光显微图像饱和伪影修复算法  

A Saturation Artifact Inpainting Algorithm for Cell Fluorescence Microscopic Images

作　　者：刘纪红[1] 张律恒 杨海旭 LIU Ji-hong;ZHANG Lü-heng;YANG Hai-xu(School of Information Science&Engineering,Northeastern University,Shenyang 110819,China;Department of Biomedical Engineering,Zhejiang University,Hangzhou 310027,China)

机构地区：[1]东北大学信息科学与工程学院,辽宁沈阳110819 [2]浙江大学生物医学工程系,浙江杭州310027

出　　处：《东北大学学报（自然科学版）》2024年第7期921-927,共7页Journal of Northeastern University(Natural Science)

摘　　要：细胞的荧光显微镜图像包含丰富的表型特征,这些表型特征被用于研究细胞内物质的吸收和运输、化学物质的分布和定位等.这些分析需要高质量的细胞图像,然而饱和伪影会导致生物表型特征的严重丢失,这将影响形态学分析和一些分类实验.从数据的后处理角度出发,提出了一种基于生成对抗网络的两阶段细胞图像修复模型,以解决饱和伪影导致的表型特征丢失问题.该模型能够修复大面积的缺失表型特征区域.通过4组实验测试了修复图像的有效性和可信度.结果表明,修复结果填补了丢失的表型特征,提高了分析中的图像质量.使用分类实验作为细胞形态学分析实验的代表,对修复前后的细胞图像进行分类实验,结果证明修复饱和伪影后的图像可以提高基于细胞形态分析的实验准确性.Fluorescence microscope images of cells contain abundant phenotypic features,which are used to study the absorption and transportation of substances in cells,as well as chemical distribution and localization.These analyses require high-quality cell images.However,saturation artifacts will cause serious loss of phenotypic features,which will affect morphological analysis and certain classification experiments.From the perspective of data post-processing,a two-stage cell image inpainting model is proposed based on generative adversarial networks to solve the loss of phenotypic features caused by saturation artifacts.The model can restore large areas of missing phenotypic characteristics.The effectiveness and reliability of the restored image are validated through four groups of experiments.The results indicate that the restored results effectively fill in the missing phenotypic features and enhance the image quality for analysis.Classification experiments,serving as a representation of cell morphology analysis experiments,are conducted on both the original and restored cell images.It is proved that the image after restoring saturated artifacts can improve the experimental accuracy based on cell morphology analysis.

关 键 词：荧光显微图像 饱和伪影 图像修复 深度学习 生成对抗网络 

分 类 号：TP391.7[自动化与计算机技术—计算机应用技术]