A Model-Based Iterative Learning Approach for Diffuse Optical Tomography.

Area of Science:

• Biomedical Optics
• Medical Imaging
• Computational Science

Background:

• Diffuse Optical Tomography (DOT) reconstructs optical properties using near-infrared light, facing challenges due to its ill-posed inverse problem.
• Image reconstruction in DOT is sensitive to measurement and modeling errors, often addressed by Bayesian methods incorporating prior information.
• Deep learning presents a promising avenue for enhancing tomographic reconstruction accuracy and efficiency.

Purpose of the Study:

• To develop and validate a model-based deep learning approach for non-linear DOT inverse problems.
• To improve the estimation of absolute absorption and scattering coefficients in DOT imaging.
• To enhance computational efficiency and address modeling errors in DOT reconstruction.

Main Methods:

• A 'model-based' deep learning strategy was employed, integrating learned components with DOT's physical model equations.
• The approach was validated using both 2D simulations and 3D experimental data.
• The method focused on estimating absolute absorption and scattering coefficients.

Main Results:

• The proposed approach demonstrated improved absorption and scattering estimates, particularly for targets with complex (smooth and sharp) features.
• It effectively compensated for modeling errors arising from coarse discretization, enabling computationally efficient solutions.
• The method achieved faster computation times compared to standard Gauss-Newton iterative methods.

Conclusions:

• Model-based deep learning offers a powerful framework for advancing DOT image reconstruction.
• This approach can accurately capture intricate image features and mitigate modeling inaccuracies.
• The technique holds potential for more efficient and accurate DOT applications in biomedical imaging.