Versatile adaptive optics microscopy through embedded neural network control

Wavefront-sensorless adaptive optics (AO) is commonly used to enable aberration estimation and correction using the information in images. We have introduced a machine learning (ML) approach that embeds physical understanding of the imaging process into a sensorless AO method. This enabled correction of aberrations with as few as two sample exposures across different microscope modalities. We extend the capabilities of such systems to more challenging imaging applications, including larger and more complex aberrations, lower signal levels, and specimen variations. We present a concept that permits estimation of multiple aberrations modes from a single image.