A 3-sided Pyramid Wavefront Sensor Controlled by a Neural Network for Adaptive Optics to reach diffraction-limited Imaging of the Retina

In-vivo imaging of the eye's fundus is the basis for the diagnosis of retinal diseases in ophthalmology. However, the resolution of the images is limited - besides the inevitable diffractive limitation by the finite size of the pupil - by the imperfect optical elements of the eye, causing aberrations. In astronomy, first test measurements of atmospheric aberrations with the novel 4-sided pyramid sensor have confirmed the benefits of this sensor compared with conventional wavefront sensors. This work presents a pyramid wavefront sensor for the measurement of aberrations and their compensation on a confocal laser scanning ophthalmoscope. By contrast, a 3-sided pyramid prism is used and it is demonstrated experimentally, that this prism, which is easier to manufacture than the 4-sided prism, reveals to be an equivalent sensor. The control of the deformable mirror which can compensate the aberrations is implemented by an artificial neural network based on the acquired sensor data. With the trained neural network, aberrations of 150nm RMS could already be reduced to half the error. The neural network is composed of a convolutional S_C-net of 2x4 layers, which extracts feature information out of the signal images of the pyramid, and a 3-layer, feed-forward backpropagation net, that determines the required deflection of the mirror in order to compensate for a given aberration.

Level thesis:degree_level

thesis.doctoral

Grantor dc:publisher

Universität Heidelberg

Year

2006

Author dc:creator

• Alvarez Diez, Cristina

Contributors dc:contributor

• Dr. Josef Bille

Repository record source_url

http://www.ub.uni-heidelberg.de/archiv/6944

OAI identifier oai:identifier

oai:archiv.ub.uni-heidelberg.de:6944