A PINN-driven game-theoretic framework in limited data photoacoustic tomography

Area of Science:

• Medical Imaging
• Computational Imaging
• Inverse Problems

Background:

• Limited data photoacoustic tomography (LD-PAT) poses reconstruction challenges.
• Accurate image reconstruction is crucial for diagnostic applications.

Purpose of the Study:

• To develop a novel framework for superior reconstructions in LD-PAT.
• To address data incompleteness using game theory and advanced computational methods.

Main Methods:

• Exploiting Cauchy data on accessible observation domains.
• Employing a Nash game-theoretic framework to complete missing data.
• Combining a gradient-free sequential quadratic Hamiltonian scheme with physics-informed neural networks.

Main Results:

• Demonstrated effectiveness in numerical simulations with various phantoms and noise levels.
• Achieved high contrast and resolution reconstructions.
• Validated robustness across different accessible observation domains.

Conclusions:

• The proposed framework offers a robust and accurate solution for LD-PAT.
• This method significantly improves reconstruction quality in limited data scenarios.
• The integration of game theory and neural networks shows promise for advanced imaging.