Deep Learning Assisted Zonal Adaptive Aberration Correction

Deep learning (DL) has been recently applied to adaptive optics (AO) to correct optical aberrations rapidly in biomedical imaging. However, because most of these methods are based on Zernike modes, they can only correct simple aberrations. Here we propose a DL assisted zonal adaptive correction method to conveniently perform corrections of high degrees of freedom while maintaining the fast speed. With a trained DL neural network, the pattern on the correction device which is divided into multiple zone phase elements can be directly inferred from the aberration distorted point-spread function image in this method. The inference can be completed in 12.6 ms with the average mean square error 0.88 when 224 zones are used. The results show the outstanding ability on aberration correction. Compared with previous method, it presents much better performance on aberrations of different complexities. Since no extra device is needed, this method has great potentials in deep tissue imaging and large volume imaging.