Adaptive optical elements & Machine Learning - The road to smart microscopy

Adaptive optical elements (AOE) such as adaptive lenses and prisms enable fast and flexible 3D-scanning in laser scanning microscopy, without the need for any mechanical movement. More complex devices enable additional functionalities, such as the correction of systematic and sample induced aberrations. One challenge in employing AOE is the control strategy, as it is not enough to simply control the input voltages, but the AOE-behaviour strongly depends on environmental conditions such as temperature and ambient pressure. To overcome these drawbacks, the behaviour of the AOE can be measured and iteratively optimized. Trained models of Machine Learning are faster than iterative strategies and need less hardware than methods based on wavefront measurements which end up in bulky and complex setups. Moreover important but unknown dependencies in huge amounts of data are detected and used as features to control the AOE properly. In this contribution, we will discuss approaches employing deep neural networks to enable precise and fast optical measurements using AOE. The combination of advanced AOE with artificial intelligence opens the door to smart microscopy.