Deep proximal gradient network for absorption coefficient recovery in photoacoustic tomography

Area of Science:

• Biomedical Optics
• Medical Imaging
• Computational Science

Background:

• Photoacoustic (PA) tomography quantifies biological tissue optical absorption via acoustic measurements.
• Conventional inversion methods are computationally intensive and sensitive to model accuracy and data completeness.

Purpose of the Study:

• Introduce a novel learned iterative method for recovering optical absorption coefficients (OACs) from PA pressure measurements.
• Address limitations of traditional iterative optimization techniques in PA tomography.

Main Methods:

• Developed a deep learning framework utilizing proximal gradient descent for optical inversion.
• Employed cascaded structural units for iterative updates of OACs through a learning process.

Main Results:

• Validated through simulations, phantom, and in vivo studies.
• Demonstrated superior performance over traditional and existing learning-based methods.
• Achieved significant improvements in relative errors, peak signal-to-noise ratios, and structural similarity.

Conclusions:

• The proposed learned iterative method enhances accuracy and efficiency in quantitative PA tomography.
• Offers a more reliable alternative to conventional methods by mitigating computational demands and sensitivity issues.
• Potential applications in advanced medical imaging and diagnostics.