Enhancement of Acoustic Microscopy Lateral Resolution: A Comparison Between Deep Learning and Two Deconvolution...

Area of Science:

• Biomedical imaging
• Microscopy techniques
• Artificial intelligence in science

Background:

• Scanning acoustic microscopy (SAM) offers high-resolution imaging of biological tissues.
• Higher transducer frequencies in SAM improve resolution but reduce penetration depth.
• Image resolution enhancement is crucial for balancing resolution and penetration in SAM.

Purpose of the Study:

• To investigate the use of deep learning (DL) for improving the lateral resolution of 180-MHz SAM images.
• To compare DL-based resolution enhancement with traditional deconvolution methods.
• To quantitatively evaluate the DL approach using higher-frequency (316-MHz) SAM images as ground truth.

Main Methods:

• Deep learning models were trained to enhance the lateral resolution of 180-MHz SAM images.
• Performance was compared against two deconvolution-based image enhancement techniques.
• Mouse and rat brain tissue sections were used as biological samples.
• Quantitative evaluation utilized 316-MHz SAM images as the ground truth.

Main Results:

• Deep learning significantly improved the lateral resolution of 180-MHz SAM images.
• DL performance closely approximated the ground truth images (NRMSE = 0.056, PSNR = 28.4 dB).
• Effective resolution enhancement was achieved even with a small training dataset (four images < 1 mm ×1 mm).

Conclusions:

• Deep learning shows high potential as a standalone superresolution method for SAM.
• DL offers a viable solution for enhancing image resolution without sacrificing penetration depth in SAM.
• This technique could advance high-resolution, deep-tissue imaging in biological and medical research.