Enhanced resolution and sensitivity acoustic-resolution photoacoustic microscopy with semi/unsupervised GANs

Area of Science:

• Biomedical optics
• Medical imaging
• Computational imaging

Background:

• Acoustic-resolution photoacoustic microscopy (AR-PAM) offers deep tissue visualization with high optical contrast.
• AR-PAM's lateral resolution and sensitivity are limited by its acoustic focusing mechanism compared to optical-resolution PAM (OR-PAM).

Purpose of the Study:

• To develop a computational strategy using generative adversarial networks (GANs) to improve AR-PAM's resolution and sensitivity.
• To enable label-free, high-resolution, and sensitive AR-PAM imaging at enhanced depths.

Main Methods:

• Employed two types of GANs: a conditional GAN for semi-supervised learning and CycleGAN for unsupervised learning.
• Utilized paired and unpaired b-scan photoacoustic microscopy (PAM) images for training GANs.
• Tested the GANs on phantom and in vivo mouse ear models.

Main Results:

• Successfully improved resolution and sensitivity in AR-PAM imaging of phantoms and in vivo mouse ears.
• Demonstrated the capability of GANs to enhance deep tissue imaging quality without requiring ground truth data.
• Achieved label-free reconstruction of AR-PAM images with enhanced resolution and sensitivity.

Conclusions:

• Generative adversarial networks offer a powerful computational approach to overcome the resolution and sensitivity limitations of AR-PAM.
• The developed semi/unsupervised GAN strategy enables high-performance AR-PAM for deep tissue imaging.
• This method holds promise for advancing biomedical visualization and diagnostics.