Deep learning estimation of eye aberrations from simulated double-pass retinal images

Konstantinos Ntatsis

Keywords: Vision and Learning, AI for Health

Abstract:

Image-based wavefront sensing allows determining optical aberrations directly via the Point Spread Function (PSF) without specialized sensors. In contrast to astronomical measurements, ocular measurements are conducted using a double-pass method. We develop a deep learning model to estimate 12 Zernike terms in the wave aberrations (excluding piston and tilt) matching each PSF. After training, the model achieves low Root Mean Square Error for the Zernike coefficients with fast one-shot inference of 3ms. These results from simulations serve as base for future work generalized to experimental data.

Link to book of abstracts (p. 62-64): https://www.vpoptics.info/past-editions/vpo24-book-of-abstracts/

Submission Number: 120