Pathological Whole Slide Hyperspectral Imaging Method And System Design

Hyperspectral imaging technology has been widely used in the field of remote sensing since it was developed.It not only contains the information of traditional color images,but also provides the imaging information of other bands,which makes the research of image analysis break through the limitation of three channel image.In the field of microscopic pathology,hyperspectral images’ abundant spectral and spatial information can provide new ideas for pathological diagnosis and solve the difficult problems that cannot be solved by traditional medical imaging methods.With the vigorous development of computer technology and automation technology,micro whole-slide scanner has been widely used.However,current whole-slide scanners are mainly designed for color imaging or fluorescence imaging,while hyperspectral imaging technology is not yet maturely applied in whole-slide imaging.This thesis proposes a complete solution of the hyperspectral whole-slide scanning and imaging system and studies the key problems in the whole-slide scanning process,mainly including whole-slide focusing method and preprocessing for image stitching.Based on these methods,this thesis realizes the micropathology hyperspectral wholeslide scanning and imaging system,which can provide research data for researchers in the field of micropathology.To provide accurate positioning information for the subsequent scanning process,this thesis proposes a tissue region recognition method,i.e.,CYU-Net,which is a twostage accurate positioning segmentation method.Then the pre-focus points are assigned on the area to be scanned and accurately focused.Because of the low speed of the traditional sharpness search based auto-focus scheme,this thesis proposes a hybrid auto-focus method based on sharpness peak search and Dv Af Net,which can predict the focus plane faster for large defocusing amount.In the process of image stitching,in order to eliminate the influence of uneven distribution of light sources and noise,an image preprocessing method based on the image content is proposed,which mainly includes blank field detection and hyperspectral image deblurring.For the purpose of screening out the blank field of view,this thesis adopts the detection method based on block Shufflenet-v2.In view of the focusing difference among different wavelengths of a hyperspectral image,a hyperspectral image deblurring enhancement method,namely HI-Net,is proposed to obtain high-quality hyperspectral images.Based on the above processing methods,this thesis finally realizes the automatic scanning and generation of whole-slide images for both hyperspectral and color filed.The experimental results show that the micropathology hyperspectral whole-slide scanning and imaging system solution proposed in this thesis has the characteristics of modular design of hardware,software scalability and good practicability of algorithm.This imaging system can output color images and hyperspectral images in single-fieldof-view or whole-slide range,which can provide a richer data source for biological tissue analysis.