Sensing Through Tissues Using Diffuse Optical Imaging and Genetic Programming

Area of Science:

• Biomedical optics
• Medical imaging
• Artificial intelligence

Background:

• Diffuse optical imaging (DOI) faces challenges in clinical adoption due to the complex inverse problem of light scattering in biological tissues.
• Existing reconstruction algorithms struggle with nonlinearity and ill-posedness, limiting DOI's application in areas like breast and brain imaging.
• Limited datasets and physical constraints further hinder the widespread use of DOI.

Purpose of the Study:

• To introduce Diffuse Optical Imaging using Genetic Programming (DI-GP), a physics-guided and interpretable AI framework for DOI.
• To develop a method that addresses the nonlinear, ill-posed inverse problem in diffuse media reconstruction.
• To enhance the speed, accuracy, and interpretability of DOI reconstructions.

Main Methods:

• Developed DI-GP, a genetic programming framework grounded in the diffusion equation.
• Evolved closed-form symbolic mappings for 2-D reconstructions in scattering media.
• Validated the approach using simulations and tabletop experiments in tissue-like media.

Main Results:

• DI-GP achieved substantially faster inference and improved reconstruction performance compared to analytical methods.
• Successfully recovered targets at depths exceeding 25 transport mean-free paths without prior knowledge of shape or location.
• Demonstrated centimeter-scale imaging in tissue-like media, showcasing potential for deep-tissue visualization.

Conclusions:

• DI-GP offers a physics-guided, interpretable, and efficient solution for diffuse optical imaging.
• The framework's transparency and data efficiency make it suitable for regulated domains requiring explainable AI.
• DI-GP shows significant promise for advancing non-invasive deep-tissue imaging and practical DOI systems.