Deep learning in photoacoustic imaging: a review

Area of Science:

• Biomedical imaging
• Medical technology
• Artificial intelligence in medicine

Background:

• Photoacoustic imaging (PAI) offers structural, functional, and molecular insights for research and clinical applications.
• Challenges in quantitative PAI (QPAI) stem from signal detection issues and unknown light distribution in deep tissues.
• Deep learning (DL) has emerged as a powerful tool for addressing PAI limitations.

Purpose of the Study:

• To provide a comprehensive overview of DL techniques applied in PAI.
• To offer guidance on designing DL models for diverse PAI tasks.
• To summarize future challenges and opportunities in DL for PAI.

Main Methods:

• A review of PAI literature focusing on DL applications published before November 2020.
• Categorization of DL applications into image understanding, initial pressure distribution reconstruction, and QPAI.
• Analysis of DL's impact on image processing, reconstruction, information fusion, and quantitative analysis in PAI.

Main Results:

• DL effectively processes PAI data, leading to enhanced image quality and reconstruction.
• DL facilitates information fusion and assists in quantitative analysis for PAI.
• DL applications span image understanding, reconstruction, and quantitative PAI.

Conclusions:

• Deep learning is a transformative tool in photoacoustic imaging.
• Continued advancements in DL theory and technology will further enhance PAI performance.
• DL is expected to accelerate the clinical adoption of PAI.