High-resolution photoacoustic microscopy with deep penetration through learning.

Area of Science:

• Biomedical Optics
• Medical Imaging
• Deep Learning

Background:

• Optical-resolution photoacoustic microscopy (OR-PAM) offers high spatial resolution but is limited to shallow imaging depths due to light scattering in tissues.
• Acoustic-resolution photoacoustic microscopy (AR-PAM) can image deeper but has lower spatial resolution.
• Bridging this gap is crucial for advanced biomedical applications.

Purpose of the Study:

• To develop a deep learning-based method for enhancing AR-PAM images.
• To achieve deep-penetrating imaging with spatial resolution comparable to OR-PAM.
• To overcome the depth limitations of OR-PAM in biological tissue imaging.

Main Methods:

• A generative adversarial network (GAN) was trained to transform blurry AR-PAM images into high-resolution images.
• The deep learning model was trained using vascular images from living mice.
• Performance was evaluated on in vivo mouse ear and brain data, comparing against blind deconvolution.

Main Results:

• The GAN significantly improved AR-PAM's lateral resolution from 54.0 µm to 5.1 µm, approaching OR-PAM's resolution (4.7 µm).
• The method produced superior microvasculature images in living mouse ears compared to blind deconvolution.
• High resolution was maintained at depths beyond one optical transport mean free path in mouse tissues.

Conclusions:

• Deep learning enables AR-PAM to achieve OR-PAM-level resolution for deep-tissue imaging.
• The proposed method overcomes the shallow imaging limitation of OR-PAM.
• This approach expands the potential applications of photoacoustic microscopy in biomedicine.