Digital refocusing in acoustic microscopy images using adversarial autoencoders

Scanning acoustic microscopy (SAM) is a label-free imaging technique that can visualize the surface and sub-surface structure and is extensively used in biomedical imaging, nondestructive testing, and material research. One of the major difficulties in high-frequency imaging is the ability to focus on the samples. This is because the ultrasound signals exhibit diffraction, which reduces the lateral resolution of the resulting images. For a better visual representation of the acquired images, the reduced lateral resolution in the out-of-focus region needs to be addressed. The synthetic aperture focusing technique (SAFT), which was developed in the 1970’s, is considered one of the most versatile techniques in acoustic imaging for addressing the issue of focusing. Despite its versatility, SAFT has some limitations. For instance, it can produce line artifacts in the reconstructed image and necessitates a defocusing distance between the sample and the focal point. Keeping these drawbacks in mind, we proposed a novel physics-based deep learning (DL) approach to effectively refocus the out-of-focus images (in this study, 4 mm).