Two-stage deep learning method for sparse-view fluorescence molecular tomography reconstruction.

Area of Science:

• Preclinical optical imaging
• Molecular imaging
• Biomedical engineering

Background:

• Fluorescence Molecular Tomography (FMT) enables cellular/molecular level tracing of physiological and pathological processes.
• Reducing FMT projection views enhances data acquisition speed, crucial for dynamic studies.
• Fewer views exacerbate the ill-posed nature of FMT inverse problems, degrading image quality.

Purpose of the Study:

• To develop a deep learning-based reconstruction method for sparse-view FMT.
• To enable high-speed FMT imaging using only four perpendicular projection views.
• To address the image degradation caused by reduced projection views in FMT.

Main Methods:

• A two-stage deep learning approach was proposed for sparse-view FMT reconstruction.
• Stage 1: A fully convolutional neural network restores surface fluorescence projection views, mitigating photon diffusion blur.
• Stage 2: A convolutional neural network performs the inverse Radon transform, reconstructing transverse slices from restored projections.

Main Results:

• The proposed deep learning method effectively reconstructs images from sparse FMT projection views.
• Numerical simulations, phantom experiments, and mouse studies validated the method's efficacy.
• The technique successfully addresses image reconstruction challenges in sparse-view FMT.

Conclusions:

• The developed deep learning method provides an effective solution for sparse-view FMT reconstruction.
• This approach significantly improves imaging speed by reducing projection views.
• The method holds promise for advancing dynamic preclinical molecular imaging applications.