Intelligent skin-removal photoacoustic computed tomography for human based on deep learning

Area of Science:

• Biomedical Imaging
• Medical Physics
• Deep Learning

Background:

• Photoacoustic computed tomography (PACT) offers centimeter-level imaging for human body detection.
• Strong photoacoustic signals from superficial skin obscure deep tissue details, limiting frontal analysis.
• Current methods struggle with artifacts and uneven illumination, hindering deep tissue visualization.

Purpose of the Study:

• To develop an automated skin removal technique for PACT imaging.
• To improve the visualization and analysis of deep tissue structures.
• To enhance the signal-to-noise ratio and reduce reconstruction errors in PACT images.

Main Methods:

• A 2.5D deep learning model utilizing a feature pyramid structure was employed.
• Single-type skin annotation was used for skin region extraction.
• A mask generation algorithm was designed for automatic skin removal.

Main Results:

• The proposed skin-removal method was validated through PACT imaging experiments on human peripheral blood vessels.
• The method demonstrated robustness against uneven illumination, irregular skin boundaries, and reconstruction artifacts.
• Reconstruction errors decreased by 20%-90%, with a 1.65 dB improvement in signal-to-noise ratio.

Conclusions:

• The developed deep learning approach effectively removes skin from PACT images.
• This technique facilitates high-definition imaging and analysis of deep tissues.
• The method offers a promising solution for overcoming limitations in current PACT imaging.