Research On Adaptive Aberrations Correction Method Of Image Scanning Microscopy Based On Deep Learning

Confocal microscopy is widely used in the field of biological microscopy due to its high resolution and three-dimensional tomography.Image scanning microscopy system replaces the point detector of a confocal microscope by an array detector,and use the algorithm instead of the pinhole to modulate the optical transfer function.The imaging resolution can reach ideal diffraction limit of 2 times and no light can damage.It is a potential low-cost super resolution stereo microscopy.However,the aberration caused by the non-uniform refractive index of biological sample and the mismatch of refractive index between sample,immersion liquid and tissue liquid will reduce the resolution and contrast sharply,which limits its application in the imaging of biological tissue and organs.The aberration caused by refractive index mismatch will change with the variation of the sample and the observation environment,so adaptive optics should be used for dynamic correction.The traditional direct wavefront detection adaptive optics method needs to introduce expensive wavefront detection unit,which makes the system structure complex,the cost is high and the optical power requirement is high.The indirect wavefront detection adaptive optics method has a long time of aberration correction,and the imaging speed of image scanning microscopy itself is slow,so the fluorescent biological samples are affected by photobleaching and phototoxicity,which leads to cell damage and incapacity.The project "Research on Adaptive Aberrations Correction Method for Image Scanning Microscopy Based on Deep Learning" aimed at the time-consuming problem of the adaptive aberration correction method for indirect wavefront detection in image scanning microimaging,the following research works were carried out:Firstly,the reconstruction algorithm of image scanning microscopy is studied and the influence of aberration is analyzed.Based on the diffraction theory,the point spread function of image scanning microscopy is constructed to reveal the inherent mechanism of pixel-reassignment and deconvolution methods to improve imaging resolution,and the simulation is carried out to verify it.Then,based on Zernike polynomial,a theoretical model of aberration was established to analyze the influence of aberration on the process of pixel redistribution and deconvolution.The single layer and multi layers media refractive index mismatch aberration models are constructed to discuss the influence of different types of aberrations on image scanning microscopy,which lays a theoretical foundation for adaptive correction of aberrations.Secondly,aberration prediction method based on deep learning is researched.The light spot image of the uniform fluorescent signal of the sample was used as the input of the deep learning aberration prediction model,and the convolution neural network was constructed to predict the aberration coefficient,and model training and aberration prediction effect analysis were carried out.The simulation results show that the prediction error of aberration coefficient i s less than 10%,which can realize effective compensation of primary aberration.Finally,the design and experimental verification of the adaptive intelligent correction system for image scanning microscopy aberration are carried out.Based on the classical image scanning microscopy system,a spatial light modulator is introduced for adaptive aberrations correction.The main aberrations types of the current system are obtained by the aberration prediction model,and then the corresponding aberrations are corrected by combining with the modal method.The experimental results show that the correction time of the proposed method is 6 2.5%shorter than of the traditional modal method.