3D k-space reflectance fluorescence tomography via deep learning.

Area of Science:

• Biomedical Optics
• Medical Imaging
• Computational Science

Background:

• Fluorescence tomography (FT) is crucial for biomedical imaging.
• Accurate 3D image reconstruction in FT remains challenging.
• Deep learning (DL) offers potential solutions for complex inverse problems.

Purpose of the Study:

• To investigate the feasibility of 3D k-space reflectance FT image reconstruction using DL.
• To develop and validate a DL model for enhanced FT imaging.
• To compare the performance of the DL approach against traditional methods.

Main Methods:

• A modified AUTOMAP deep learning architecture was employed.
• An open-source Monte Carlo simulator generated training data.
• An enhanced EMNIST dataset served as an embedded contrast function.
• k-space illumination in a reflectance configuration was utilized.
• In silico and phantom experiments validated the DL model.

Main Results:

• The DL model successfully reconstructed single and multiple fluorescent embeddings in 3D.
• The proposed DL technique outperformed least-squares (LSQ) and total-variation minimization (TVAL) methods.
• Superior performance was particularly noted at greater depths.

Conclusions:

• Deep learning provides a powerful computational tool for 3D k-space reflectance FT.
• The developed DL approach offers improved accuracy and depth penetration compared to traditional methods.
• This technique holds significant potential for future preclinical imaging applications.