Machine-learning methods for adaptive optical microscopy

Optical microscopy is commonly used for non-invasive imaging of biological structures at subcellular resolution. However, in complex specimens, image quality can be compromised by optical aberrations caused by the sample. While adaptive optics can correct these distortions and restore diffraction-limited resolution, they require wavefront sensing and corrective devices, which adds to the system's complexity and cost. I will discuss recently developed machine learning algorithms that perform simultaneous wavefront estimation and three-dimensional structural information extraction from a single three-dimensional image stack, without the need for external training datasets. Additionally, I will provide examples of how these algorithms are applied to in vivo imaging of brain structure and function.