Machine learning based adaptive optics for doughnut-shaped beam

Abstract

The doughnut-shaped beam has been widely applied in the field of super-resolution microscopic imaging, micro-nanostructure lithography, ultra-high-density storage, and laser trapping. However, how to maintain the doughnut-shaped focus inside the scattering medium becomes a challenge, due to the wavefront aberrations. Here we demonstrate a machine learning based adaptive optics method to recover the doughnut-shaped focus with high speed. In our method, the relationship between the distorted doughnut-shaped intensity point spread function and the coefficients of the first 15 Zernike modes for phase correction is established. Experimental results show that the wavefront aberration with 101,784 optical control elements can be predicted within ~17 ms even using a personal computer, and 97.5% correction accuracy can be achieved in 200 repeated tests. Besides, we successfully apply this method in the scanning microscopy theoretically. With a large number of optical control elements and fast operation speed, our method may pave the way for many important applications in bioimaging, such as deep tissue stimulated emission depletion (STED) microscopy.