Research On Image Scanning Microscopy Imaging Method Based On Generative Adversarial Networks

Image Scanning Microscopy(ISM)imaging method has high resolution imaging characteristics and has broad application prospects in the biomedical field.ISM can achieve high resolution imaging by using array scanning method on the basis of confocal microscopy,which can improve the resolution to twice the diffraction limit.The experimental device is simple and easy to implement,which has become a research hotspot in the field of super-resolution microscopy.However,this method needs to collect a large number of images,and the data processing is large and the imaging speed is slow.Due to the limitation of scanning range and objective numerical aperture,it cannot meet the needs of large field of view fast imaging.The subject “Research on image scanning microscopic imaging method based on generative adversarial network” aims at the problem that image scanning microscopic imaging is difficult to realize large field of view and fast imaging,and carries out the following research work:(1)The ISM imaging method based on Fourier reweighted is studied.The comparative study of confocal scanning microscopy and image scanning microscopy was carried out,and the theoretical model of ISM imaging method based on pixel redistribution was established,revealing the internal mechanism of imaging resolution improvement.The deconvolution image reconstruction algorithm based on Fourier reweighted is studied.The ISM simulation experiment is carried out for the simulated microtubule samples.It is proved that the resolution is significantly higher than that of the confocal microscope,which can reach twice the diffraction limit.(2)An ISM image prediction method based on Generative Adversarial Network(GAN)is proposed.The actual ISM images and wide field microscopic images are input into the network,and the loss function based on multi-parameter fusion is constructed for GAN training.Based on the output GAN model parameters,the fast prediction of highresolution ISM images in wide field microscopic field of view is realized.The method has the ability of transfer learning and avoids a large number of image acquisition in the traditional ISM imaging.The high-resolution ISM image is predicted based on the wide field microscope device to solve the problems of limited field of view and slow imaging speed of ISM.(3)GAN training and prediction effect evaluation were carried out.Adam algorithm is used for GAN training and super-parameter adjustment and optimization.A multi-index evaluation system based on structural similarity,peak signal-to-noise ratio and spatial logarithmic spectrum is constructed,and the proximity between the predicted results and the real results is evaluated from three aspects of image microstructure,signal strength and information richness.The large field of view ISM prediction experiment is carried out.The ISM imaging system is built to collect images for network training.By using the trained network and selecting the wide field 3T3 cell sample image as input,the high resolution and large field of view ISM image is quickly and successfully predicted.The resolution is obviously improved.The field of view is 1 mm,and the imaging speed is only 0.4 s slower than that of the wide field imaging,which is much higher than that of the traditional ISM method.