Machine Learning and Neural Representations for Enhancing Phase Diversity-based Adaptive Optics
Magdalena Schneider, Computation & Theory Fellow at the Janelia Research Campus
Leuchs-Russell-Auditorium, A.1.500, Staudtstr. 2
Abstract:
Fluorescence microscopy often suffers from distortions caused by optical imperfections or sample inhomogeneities. Adaptive optics (AO) corrects these issues by detecting and adjusting wavefront aberrations. We demonstrated that phase diversity, a method previously used in astronomy, can be effectively adapted to fluorescence microscopy. This sensorless technique estimates aberrations from a series of images taken with controlled distortions. In our recent work, we trained a machine learning model to calibrate the deformable mirror for applying phase diversities and correcting aberrations. Leveraging neural representations of the unknown sample and phase aberration, we achieve precise reconstruction of sample structures even under severe, high-order aberrations.
