Abstract
The Point Spread Function (PSF) of the human eye is determined by both optical aberrations and straylight. However, accurately retrieving underlying wavefront aberrations from PSF images becomes challenging when straylight is present due to their combined effects in the resulting image. Traditional wavefront sensing techniques struggle to separate these contributions, limiting clinical assessment of optical quality. We propose a deep learning-based method to retrieve the underlying wavefront-aberration from simulated PSFs with straylight. The effect of scatter is implemented as an additional random phase perturbations wavefront. The model can predict the wavefront with high accuracy, achieving one-shot inference in 3 ms. This approach could enable a more comprehensive assessment of ocular optical quality by separating aberration and scatter components from standard PSF measurements.