Study On Whole Slide Imaging

Whole slide imaging(WSI)is a typical large field-of-view(FOV)and high-resolution imaging approach.The WSI uses high magnification and high numerical aperture(NA)objective lens or super-resolution recovering algorithm to acquire high-resolution image,and enables a large FOV by scanning or non-scanning style,finally forming a digital,clear,seamless and whole-slide image.Digital pathology is the fundamental stone of next-generation cancer diagnosis and analysis,and a principle implementation is thanks to the WSI.This thesis introduces the development of the WSI systematically,and give a summary of the current autofocusing technologies,as well as perform lots of solid research to overcome the scientific problems of the trade-off between FOV and resolution and the incapability of accessing the quantitative phase via two different WSI techniques based on the scanning-stitching style on a microscopy platform or computational approach via a lensless on-chip imaging system.We have proposed effective solutions to overcome the technical issues existed in the conventional microscope and commercial WSI products,including non-automatic operation,subjective image-reviewing criterion,limited depth of field,low imaging throughput,non-realtime autofocusing and simple imaging modality.The primary works and innovations are as follows:(1)To solve the technical issue of the non-realtime autofocusing in comercial WSI products,we proposed a real-time,single-shot autofocusing technique based on a color-multiplexed LED illumination and build up a high-throughput brightfield WSI prototype.The system used a programmable photographic lens to replace the tube lens that is usually configured on a conventional microscope,enabling autofocusing without changing the relative distance between sample and objective lens and performing rapid 3D sectioning imaging.We can capture the color-multiplexed image to calculate the defocus distance during the movement of the sample thanks to the use of mutual information based image registration method.We also presented a correction method to resolve the color-crosstalk problem in image acqusition.In addition,to remove the stitching errors in whole-slide image,we used a hole-array mask and standard microscope to measure the pincushion distortion in photographic lens and created a correction strategy via the data fitting.The measured resolution of the reported system is～700 nm and the averaged focusing error is～0.33μm.Our system can acquire whole slide images of a 225-mm~2 region in～2 mins.(2)To solve the technical issues of the non-realtime autofocusing and the simple imaging mode in comercial WSI products,we reported a real-time,single-shot autofocusing technique via dual-LED illumination and build up a WSI prototype with brightfield,fluorescence and phase-contrast imaging modes.Instead of mutual information,we used autocorrelation to determine the defocus distance only with a single image,enabling 83%improvement on the consumed time compared to the previous work.The reported system has averaged focusing error of～0.34μm.Besides,we proposed a totally automatic digital distortion correction strategy,and there is no need of any standard hole-array mask and microscope to measure the distortion.(3)To overcome the disadvantages of the low imaging throughput,limited depth of field,and simple imaging modality in WSI system,we proposed a virtual WSI,termed DeepWSI,based on the unsupervised deep-learning and image-to-image translation.We used the cycleGAN as the network,and used msSSIM as the loss function for the image translation.We demonstrated that images captured with a regular 10×/0.3NA objective lens can be transformed into images comparable to that of a 60×/1.4-NA oil immersion lens.We also implemented a high-throughput WSI with equivalent data throughput of>2 gigapixels per second.Furthermore,the DeepWSI virtually refocused the blurred image to the in-focus plane for extending the depth of field of the system.Finally,we also reported a high-resolution virtual staining strategy and demonstrate it for Fourier ptychographic WSI.(4)To resolve the problems of the low imaging throughput and low imaging resolution in current ptychography imaging,we reported a near-field ptychographic lensless imaging technique via the modulation of the disorder-engineered-surface.The engineered surface consists of micron-level phase scatters and sub-wavelength intensity absorbers.We also proposed a new coherent diffraction imaging model by considering both the spatial and angular responses of the pixel readouts,enabling resolving 616 nm resolution with the wavelength of 405 nm and reaching equivalent NA of～0.8.Besides,we can quantitatively recover slow-varying 3D phase objects with many 2πwraps.We also build up a high-throughput parallel ptychographic imaging system with a 240-mm~2 effective field of view,which can be acquired in 15 seconds.Finally,we performed virtual staining by using the recovered phase as attention guidance in a cycleGAN,enabling computational brightfield,fluorescence WSI.(5)To solve the issues of unsatisfied image recovery and incapability of accessing the quantitative phase of the slow-varying object,we reported a multi-height phase retrieval via a coded image sensor.The coded layer is made by coating a thin layer of microbeads on top of the image sensor.The simulation study demonstrated that the coded sensor approach performs better than that of the regular multi-height approach for both slow-and fast-varying phase objects.We also validated with several experiments using USAF resolution target,quantitative phase target,tissue slides and bacterial colony,and the experimental results were in a good agreement with the simulation.Finally,we also performed a WSI using the reported scheme.