UHR-DeepFMT: Ultra-High Spatial Resolution Reconstruction of Fluorescence Molecular Tomography Based on 3-D Fusion Dual-Sampling Deep Neural Network
- 7 April 2021
- journal article
- research article
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Medical Imaging
- Vol. 40 (11), 3217-3228
- https://doi.org/10.1109/tmi.2021.3071556
Abstract
Fluorescence molecular tomography (FMT) is a promising and high sensitivity imaging modality that can reconstruct the three-dimensional (3D) distribution of interior fluorescent sources. However, the spatial resolution of FMT has encountered an insurmountable bottleneck and cannot be substantially improved, due to the simplified forward model and the severely ill-posed inverse problem. In this work, a 3D fusion dual-sampling convolutional neural network, namely UHR-DeepFMT, was proposed to achieve ultra-high spatial resolution reconstruction of FMT. Under this framework, the UHR-DeepFMT does not need to explicitly solve the FMT forward and inverse problems. Instead, it directly establishes an end-to-end mapping model to reconstruct the fluorescent sources, which can enormously eliminate the modeling errors. Besides, a novel fusion mechanism that integrates the dual-sampling strategy and the squeeze-and-excitation (SE) module is introduced into the skip connection of UHR-DeepFMT, which can significantly improve the spatial resolution by greatly alleviating the ill-posedness of the inverse problem. To evaluate the performance of UHR-DeepFMT network model, numerical simulations, physical phantom and in vivo experiments were conducted. The results demonstrated that the proposed UHR-DeepFMT can outperform the cutting-edge methods and achieve ultra-high spatial resolution reconstruction of FMT with the powerful ability to distinguish adjacent targets with a minimal edge-to-edge distance (EED) of 0.5 mm. It is assumed that this research is a significant improvement for FMT in terms of spatial resolution and overall imaging quality, which could promote the precise diagnosis and preclinical application of small animals in the future.Funding Information
- National Key Research and Development Program of China (2017YFA0700401)
- Fundamental Research Funds for Central Universities
- National Natural Science Foundation of China (61871022, 61601019)
- Beijing Natural Science Foundation (7202102)
- Higher Education Discipline Innovation Project (B13003)
This publication has 42 references indexed in Scilit:
- High-Resolution Fluorescence Diffuse Optical Tomography Developed with Nonlinear Upconverting NanoparticlesACS Nano, 2012
- JointL1and total variation regularization for fluorescence molecular tomographyPhysics in Medicine & Biology, 2012
- MAP estimation with structural priors for fluorescence molecular tomographyPhysics in Medicine & Biology, 2012
- Noninvasive molecular imaging of c-Myc activation in living miceProceedings of the National Academy of Sciences of the United States of America, 2010
- Going deeper than microscopy: the optical imaging frontier in biologyNature Methods, 2010
- Adaptive finite element methods for the solution of inverse problems in optical tomographyInverse Problems, 2008
- FLUORESCENCE MOLECULAR IMAGINGAnnual Review of Biomedical Engineering, 2006
- Looking and listening to light: the evolution of whole-body photonic imagingNature Biotechnology, 2005
- Optical tomography in medical imagingInverse Problems, 1999
- Analysis of Discrete Ill-Posed Problems by Means of the L-CurveSIAM Review, 1992