Image Processing Algorithm For Light Sheet Microscopy

Along with the rapid development of computer science,molecular biology and medical imaging technology,molecular imaging technology has made great leap.The light sheet microscopic system(LSM),which belongs to the mesoscopic imaging modalities among various molecular imaging technologies,is suitable for imaging of the entire tissue because of the advantages of high speed imaging and low light toxicity.And its field-of-view can be 1 to 2 cm,meanwhile its resolution can reach micron level.However the laser could be absorbed,scattered and refracted in the transmission processbecause of different density and structures of the local tissue.If the transparency of the tissue is not high,there will be stripe artifacts,scattering blur and gray inhomogeneity in LSM images.In addition,the thickness of the illumination sheet and the spacing of the light sheet are much larger than the resolution of the collected image plane,which leads to an axial resolution of light sheet data much less than 10 times the lateral resolution.Improving the quality of microscopic image boosts the application of LSM in biological research.In the case of cerebral blood vessels,reducing the noise of microscopic image and extracting the blood vessel structure facilitate the qualitative and quantitative research on the causes and mechanisms of brain diseases.Therefore,it is important to study image processing algorithm for light sheet microscopy.In this paper,a series of studies are carried out on the main line of LSM,including multi-view fusion algorithm,enhancement and segmentation algorithm for LSM images.Feature extraction,matching and nearest neighbor interpolation are used to fuse the fluorescence signals with different angles and acquire more details of the tissue in multi-view fusion algorithm.In order to improve the image quality,a prior feature network structure is used to enhance the vascular structure.The proposed Parallel convolutional neural network with prior feature network(PCNN)converts the vessel segmentation into a classification for pixels,and dice similarity coefficient is 90.42%.