Research On Super Resolution Reconstruction Of Optical Projection Tomography

Based on the characteristics of nearly straight light propagation in clearing medium,optical tomography projection(OPT)technique utilizes the projections from different angles after the light beam penetrating the biological sample to reconstruct three-dimensional images of the sample.OPT technology is mainly used for imaging of 1-10 mm biological samples.It has the advantages of high resolution,low cost of equipment,convenient operation,and both structural and molecular imaging ability.It is very well suited to the needs of small scale imaging of biological samples in the field of life sciences.However,current OPT technology is limited by the imaging system structure,optical depth of field,geometric correction and sample morphology,and it is difficult to obtain high-resolution multi-angle projection data.Therefore,high-resolution reconstruction technology is needed in OPT imaging.In this thesis,super-resolution reconstruction for OPT imaging is carried out by using the in-house developed OPT imaging system to deal with the problem of insufficient resolution in OPT imaging.The main research work of this paper is as follows:1.OPT projection data acquisition and a survey of image super-resolution reconstruction methods.The projection data of a large number of drosophila pupae and Arabidopsis silique samples under different light sources were collected by using the in-house developed OPT system,which are used as the data source for the super-resolution reconstruction.Then,current image super-resolution reconstruction methods are reviewed with emphasis on several super-resolution reconstruction methods based on deep learning and the corresponding reconstruction image quality evaluation indices are introduced in details.2.Simulation study of OPT projection super-resolution reconstruction.A large number of clear projections are selected from the data collected by the OPT system.The low resolution images are obtained by down-sampling the high-resolution projection,and the mapping between the corresponding high and low resolution image blocks is acquired by the deep convolution neural network based methods and the A + method which is based on traditional machine learning.The experimental results show that the OPT images super-resolution reconstructed by using the map relationship are better than the interpolated images as indicated by the evaluation indices and the subjective visual inspection.3.Real projection data super-resolution reconstruction of the OPT system.The OPT system collects the projection data of the same sample at different magnifications,and the common region of intrest(ROI)of the projection images with the same projection angle at different magnification are registrered to obtain the corresponding blocks of the high and low resolution projections of the real system.Four super-resolution reconstruction methods based on deep convolution neural networks are used for super-resolution reconstruction.The experimental results show that the super-resolution reconstruction guided by the mapping relationship of the deep convolution networks can achieve better results than the interpolation method when the projection image quality is good enough and the registration precision is guaranteed.