Image reconstruction of multispectral sparse sampling photoacoustic tomography based on deep algorithm unrolling.

Area of Science:

• Medical Imaging
• Biomedical Engineering
• Computational Imaging

Background:

• Photoacoustic tomography (PAT) offers in vivo structural, functional, and metabolic imaging of biological tissues.
• Sparse Sampling PAT (SS-PAT) reduces detector needs but faces ill-posed image reconstruction challenges.
• Current SS-PAT reconstruction methods include model-based (requiring complex priors) and deep-learning (lacking interpretability).

Purpose of the Study:

• To develop a novel SS-PAT image reconstruction method integrating model-based and deep-learning advantages.
• To address the ill-posed nature of SS-PAT reconstruction using a deep algorithm unrolling (DAU) framework.
• To incorporate structural prior constraints for improved sparse sampling reconstruction.

Main Methods:

• Proposed a novel Deep Algorithm Unrolling (DAU) framework for SS-PAT image reconstruction.
• Developed a nested DAU framework utilizing plug-and-play Alternating Direction Method of Multipliers (PnP-ADMM) to handle sparse sampling.
• Analyzed the DAU approach for PAT reconstruction and incorporated structural prior constraints.

Main Results:

• The proposed DAU framework demonstrated superior performance in SS-PAT image reconstruction.
• Experimental results from numerical simulations, in vivo animal imaging, and multispectral un-mixing validated the method's effectiveness.
• The DAU-based reconstruction outperformed existing state-of-the-art model-based and deep-learning methods.

Conclusions:

• The novel DAU framework effectively integrates model-based and deep-learning strengths for SS-PAT.
• The PnP-ADMM-based nested DAU approach successfully addresses sparse sampling challenges in PAT.
• This method offers a promising advancement for high-quality in vivo biological tissue imaging using SS-PAT.