Research On Cerenkov Fluorescence 3D Imaging Method Based On Deep Neural Network

Cerenkov Luminescence Tomography(CLT),as an important part of Optical Molecular Imaging(OMI),has a large number of radionuclide probes that can be used in clinical practice and has received extensive attention from researchers.The basis of CLT technology is Cherenkov radiation produced by the Cherenkov effect,but because it is a secondary product of decay,the fluorescence signal intensity is low,it is susceptible to noise interference,and there is serious ill-posedness in the reconstruction process.The reconstruction result has artifacts,which seriously affects the reconstruction quality of CLT,which is far from the actual CL light source,so it cannot be widely used in the clinical field.Focusing on the above issues,this article proposes two different CLT reconstruction frameworks to improve the reconstruction results.The specific work is as follows:(1)A CLT reconstruction framework based on stack denoising autoencoder is proposed,which aims to solve the problem of poor quality of reconstruction results.The framework is a cyclic structure.In each cycle,the reconstruction area is divided into feasible and nonfeasible areas,and the stack denoising autoencoder is used to extract the features of the nearest quaternary domain of each node in the feasible domain.After that,the fuzzy C-means clustering algorithm is used to divide the extracted features to obtain new feasible and infeasible regions,and the final feasible region is the reconstruction target.This article designed a variety of simulation experiments to prove that the reconstruction framework can indeed obtain more accurate CLT reconstruction results.(2)A new method of local connection network reconfiguration based on finite element dissection information is proposed.Its purpose is to introduce the local connection relationship between nodes and correct the result of the fully connected network.The main part of the method is a fully connected network to obtain preliminary reconstruction results.Then there is a local connection network established based on the finite element dissection information to obtain the residual between the preliminary reconstruction result and the real target.Finally,a residual connection is made between the fully connected network and the partially connected network,and the final reconstruction result can be obtained.A variety of simulation experiments are designed in the article to prove the stability of the reconstruction method and good reconstruction performance.