MUnet: A model-based unrolled reconstruction framework for computational photoacoustic imaging

Area of Science:

• Medical imaging
• Biomedical engineering
• Computational imaging

Background:

• Photoacoustic imaging (PAI) offers unique optical contrast and ultrasonic penetration for in vivo biological visualization.
• Robust PAI image reconstruction is hindered by limited-view detection, sparse sampling, and noise.
• Current reconstruction methods involve trade-offs between efficiency, accuracy, and interpretability.

Purpose of the Study:

• To develop an advanced image reconstruction method for photoacoustic imaging.
• To overcome the limitations of existing PAI reconstruction techniques, particularly under challenging conditions.
• To improve the speed and fidelity of PAI for potential real-time clinical applications.

Main Methods:

• Proposed MUnet, a model-based unrolled deep learning architecture.
• Integrated a physics-informed data-fidelity term with a supervised image-domain enhancement module.
• Alternated non-trainable analytic updates with neural image refinement for enhanced reconstruction.

Main Results:

• MUnet demonstrated superior performance compared to traditional and learning-based methods on in vivo PAI data.
• Achieved significantly improved image quality and reduced computation time.
• Validated the effectiveness of combining model-based interpretability with deep learning flexibility.

Conclusions:

• MUnet offers a promising solution for rapid and high-fidelity photoacoustic imaging.
• The proposed method enhances PAI reconstruction robustness and efficiency.
• MUnet advances the potential of PAI for real-time clinical diagnostics and imaging.