A Reconstruction Method Based On Point Cloud Deep Convolution Neural Network For Bioluminescence Tomography

Bioluminescence tomography(BLT)is an optical molecular imaging technology with broad application prospects.It is non-invasive and highly sensitive,and can quantitatively monitor pathological and physiological changes of diseased areas in biological bodies at cellular and molecular levels.BLT combined with CT imaging technology simultaneously collects internal structure information of biological tissues and optical information of biological surface,and then detects and images the lesions in biological bodies.BLT has the advantages of low cost and high sensitivity,it has more and more important application potential in clinical research and disease diagnosis.However,due to the complex transmission of photons in biological tissues,the measurable optical information on the surface of the organism is insufficient,so the reconstruction of BLT has serious ill-posedness.In this paper,aiming at the limitations of the traditional sparse regular reconstruction algorithm,point cloud representation was introduced and combined with deep convolution neural network to carry out research.The main research contents of this paper are as follows:(1)Aiming at the limitation of the traditional sparse regular light source reconstruction algorithm that the morphology of the reconstructed light source depends on the density of the finite element mesh,the point cloud representation was introduced in the field of bioluminescence tomography,we proposed the deep convolution neural network model,implements the direct light source target reconstruction in three-dimensional numerical space,the dependence on finite element mesh in light source reconstruction is eliminated.The numerical simulation experiments show that the proposed reconstruction method based on point cloud integrated deep convolution neural network,Point-DCNN,achieves good target location accuracy and target shape reconstruction accuracy under various experimental settings.(2)To optimize the performance of the Point-DCNN reconstruction method,we fused prior constraint information and attention mechanism into the deep convolutional neural network model and proposed optimized siamese network Point-DCNN reconstruction method.The numerical simulation experiments show that the siamese network Point-DCNN reconstruction method with prior constraint and attention mechanism has good localization ability and morphological reconstruction ability for single light source reconstruction and double light source reconstruction.