Deep learning in macroscopic diffuse optical imaging.

Area of Science:

• Biomedical optics
• Computational modeling
• Medical imaging

Background:

• Classical physical modeling and signal processing have traditionally guided biomedical optics.
• Deep learning (DL) has emerged as a powerful computational tool across various scientific domains.
• DL offers new approaches for complex data analysis in scientific research.

Purpose of the Study:

• To comprehensively review the application of deep learning (DL) in macroscopic diffuse optical imaging (DOI).
• To provide an accessible introduction to DL for a broader audience.
• To summarize current DL research in key DOI areas.

Main Methods:

• Review of existing literature on DL in diffuse optical imaging.
• Explanation of DL fundamentals for a general audience.
• Summary of DL applications in optical properties retrieval, fluorescence lifetime imaging, and diffuse optical tomography.

Main Results:

• DL significantly reduces analysis time in DOI, often by orders of magnitude.
• DL improves quantitative reconstruction quality and robustness to noise compared to conventional methods.
• DL can learn complex end-to-end relationships, overcoming limitations of traditional inverse solvers.

Conclusions:

• DL's validated capabilities offer immense potential for novel DOI modalities.
• DL can facilitate the clinical translation of previously impractical DOI techniques.
• The integration of DL promises to bring advanced imaging solutions to patient care.