Research On Cherenkov Fluorescence Tomography Based On Improved Fluorescence Image Quality

Cerenkov Luminescence Imaging(CLI)has become a research hotspot in the field of optical molecular imaging in recent years because it has a wide range of clinically available radionuclide probes.Based on the Cherenkov effect,this technology is imaging by detecting the Cherenkov fluorescence produced by the decay of the radionuclide.However the intensity of Cherenkov fluorescence is weak,it always takes a long time to get the fluorescence signal for imaging.At the same time,because the decay process of the radionuclide is accompanied by a large number of high-energy rays,the Cherenkov fluorescence image obtained will be polluted by noise in the long process of collection.This phenomenon seriously affects the quality of the CLI image and makes it difficult on the quantitative analysis of the CLI and three-dimensional reconstruction of Cerenkov Luminescence Tomography(CLT).In order to explore the impact of CLI noise image on CLT reconstruction,this paper has conducted preliminary research on the method of CLI image denoising.The specific work is as follows:(1)In view of the image pollution caused by high-energy ray during radionuclide decay,this paper proposes a CLI image denoising method based on fuzzy local information C-means and Total Variation model.The CLI image denoising method includes two steps:segmentation of noise pixel region and image inpainting.In order to reduce the sensitivity of the unsupervised learning algorithm about the noise,the spatial local information is added to clustering algorithm.Then the classification result is mainly analyzed to image denoising and repair.The effect of image edge and detail in denoising process is reduced by specific denoising and restoration of different classification results.Results of the simulation experiments and real experiments prove that the proposed denoising method can remove the impulse noise effectively with the ability to maintain the shape of Cerenkov Luminescence source.(2)In view of the fact that unsupervised learning clustering algorithm is inevitable affected by noise in the step of noise pixel region segmentation,it interferes the classification results.Inspired by deep learning,this paper introduces a fuzzy clustering denoising method based on stack denoising Autoencoders.A neighborhood of a pixel is used as the input data of the network,and then the high-level features of the network are extracted asthe input of the clustering algorithm.Compared with directly using pixel features as input of clustering algorithm,extracting abstract features of image can further reduce the sensitivity to noise in the step of noise pixel region segmentation,so as to improving the classification performance.Considering that the CLI noise image will affect the accuracy of CLT reconstruction,this paper further explores the effect of two denoising methods on reconstruction results.Simulation experiments and physical experiments prove that the noise will reduce the accuracy of CLT reconstruction to some extent.After pre-processing with the proposed denoising method before reconstruction,the accuracy of reconstruction can be improved.