Structure-fused deep 3D hierarchical network: A bioluminescence tomography scheme for different imaging objects

Beilei Wang,Shuangchen Li,Xuelei He,Yizhe Zhao,Heng Zhang,Xiaowei He,Jingjing Yu,Hongbo Guo
DOI: https://doi.org/10.1109/EMBC40787.2023.10340967
2023-01-01
Abstract:Monte Carlo eXtreme (MCX) method has a unique advantage for deep neural network based bioluminescence tomography (BLT) reconstruction. However, this method ignores the distribution of sources energy and relies on the determined tissue structure. In this paper, a deep 3D hierarchical reconstruction network for BLT was proposed where the inputs were divided into two parts -- bioluminescence image (BLI) and anatomy of the imaged object by CT. Firstly, a parallel encoder is used to feature the original BLI & CT slices and integrate their features to distinguish the different tissue structure of imaging objects; Secondly, GRU is used to fit the spatial information of different slices and convert it into 3D features; Finally, the 3D features are decoded to the spacial and energy information of source by a symmetrical decoding structure. Our research suggested that this method can effectively compute the radiation intensity and the spatial distribution of the source for different imaging object.
What problem does this paper attempt to address?