Quantitative photoacoustic tomography based on a physics-constrained deep learning framework with implicit priors.

Area of Science:

• Biomedical Imaging
• Computational Imaging
• Medical Physics

Background:

• Quantitative photoacoustic tomography (qPAT) is challenged by the ill-posed nature of optical inverse problems, hindering accurate reconstruction of optical absorption coefficients.
• Existing methods often struggle with image quality and quantitative accuracy due to these inherent challenges.

Purpose of the Study:

• To develop a novel framework, PIDP-qPAT, that integrates physical models with implicit deep learning priors for improved qPAT reconstruction.
• To address the ill-posed nature of the optical inverse problem in qPAT without requiring specialized training data.

Main Methods:

• PIDP-qPAT utilizes a pre-trained denoising network as an implicit regularizer within an optimization framework.
• The reconstruction is decoupled into physics-driven data fidelity and learning-driven denoising sub-tasks, solved via alternating optimization.
• This approach leverages natural image priors, eliminating the need for specific training datasets.

Main Results:

• PIDP-qPAT demonstrated superior performance over state-of-the-art techniques in simulations, phantom experiments, and in vivo studies.
• The framework achieved significant improvements in both visual quality and quantitative metrics.
• Computational efficiency was comparable to purely data-driven methods.

Conclusions:

• PIDP-qPAT offers a robust and practical solution for quantitative photoacoustic tomography.
• The integration of physical models with data-driven priors effectively addresses ill-posed inverse problems in imaging.
• This approach has broad applicability to other ill-posed imaging modalities beyond qPAT.