Research On Aberration Correction Methods Based On Wavefront Detection And Deep Learning In Microscopy

Fluorescence microscopy plays an important role in biomedical research.However,the image quality of the microscopy is limited by the aberrations of the microscopy itself and the aberrations introduced by the biological tissues.Adaptive optics can measure and correct the aberrations accordingly,which enables the microscopy to effectively focus the light deep in tissue,so as to achieve the deep tissue focusing and recover diffraction-limited imaging performance.However,the wavefront detection and aberration correction methods in microscopy are still worthy of further research and improvement.When using the wavefront detection based aberration correction methods in microscopy to focus light deep in biological tissue,the focusing reliability requires further investigation.When Shack-Hartmann wavefront sensor is used for wavefront detection,the detection process needs to be further simplified,and the detection accuracy and speed could be further improved.When there is no adaptive optics hardware support in microscopy,it is necessary to develope the hardware independent aberration correction method as an alternative.In this dissertation,the aberration correction methods based on wavefront detection and deep learning in microscopy are studied in depth.The main contents and innovations of this dissertation are as follows:(1)The wavefront detection reliability of aberration correction method based on coherent optical adaptive technique using backscattered light as feedback on mouse brain slices is investigated experimentally.It is proposed that the radius of the Airy disk is used as the focusing reliability criterion.The critical depth of reliable focusing of coherent optical adaptive technique in this experimental condition is 150?m.The experimental results fill the gap of coherent optical adaptive technique in the research of wavefront detection reliability,and provide experience support for its application in biomedical reserach.(2)A wavefront detection method based on deep learning is proposed to predict the Zernike coefficients directly from the images acquired by the Shack-Hartmann wavefront sensor through convolution neural network.The proposed method simplifies the process of wavefront detection and improves the accuracy and speed of wavefront detection.Experiments on 300?m thick mouse brain slice verify the performance of this method on wavefront detection and aberration correction.(3)A wavefront reconstruction method based on deep learning is proposed to reconstruct the wavefront distribution directly from the image acquired by the Shack-Hartmann wavefront sensor,which further simplifies the process of wavefront detection and improves the detection accuracy.Compared with other wavefront reconstruction methods,the performance and generalization ability of the proposed method in wavefront reconstruction and aberration correction are verified.(4)Based on deep learning and prior knowledge of aberration,an aberration correction method for image restoration by convolution neural network is developed for the microscopy without adaptive optics hardware,which can achieve the image aberration correction and noise reduction in high speed and does not require wavefront detection.A datasets generation pipeline is proposed to generate and augment the training datasets from a small number of acquired images,which does not require accurate image registration,and can save the labor cost and time consumption of data acquisition.The proposed method has verified the performance of image restoration in several commercial fluorescence microscopes and a custom-made fluorescence microscope,and shows good generalization ability.In summary,this dissertation experimentally investigates the wavefront detection reliability of aberration correction method based on coherent optical adaptive technique in microscopy,and the aberration correction methods based on wavefront detection,deep learning and image restoration are proposed,which have the potential application prospects in biomedical research.