STED microscopy with two-photon photoactivation in the visible range & Application of generative adversarial networks for image reconstruction in STED microscopy
by Jan-Erik Bredfeldt
Date of Examination:2022-12-05
Date of issue:2023-01-31
Advisor:Prof. Dr. Stefan Hell
Referee:Prof. Dr. Stefan Hell
Referee:Prof. Dr. Tim Salditt
Referee:Prof. Dr. Claus Ropers
Referee:Prof. Dr. Alexander Egner
Referee:Prof. Dr. Stefan Jakobs
Referee:Prof. Dr. Silvio Rizzoli
Files in this item
This file will be freely accessible after 2023-12-04.
EnglishThe resolution limit of classical optical far-ﬁeld microscopes was broken with the invention of super-resolution microscopy. Today, a variety of these techniques achieving diﬀerent degrees of resolution exist. Image formation with most of these methods can be easily aﬀected by either photobleaching of ﬂuorescent dyes due to the use of high-intensity lasers or out-of-focus ﬂuorescence signal from dyes in diﬀerent imaging planes. The resulting loss in contrast reduces the possibility to distinguish individual structures or to localize individual ﬂuorophores, ultimately limiting the gain in resolution. This is especially the case when imaging axially extended, densely labeled biological samples, where the out-of-focus signal can be quite substantial. Two-photon excitation as well as two-photon photoactivation (2PA) are promising strategies to mitigate this problem. The sharply conﬁned two-photon active volume facilitates imaging with signiﬁcantly reduced background signal. The use of STED compatible photoactivatable dyes enables the application of leading-edge imaging modes including the nanometer-resolution MINSTED nanoscopy. The goal of this thesis is to present eﬃcient 2PA in the visible spectrum with multiple photoactivatable dyes. The broad application of this technique, together with its beneﬁts compared to regular one-photon activation, is demonstrated for diﬀerent microscopy techniques. On the other hand, in cases without 2PA, a deteriorated signal-to-background ratio in measurements that are aﬀected by a loss in contrast can be potentially recovered through post-processing. The second part of this thesis investigates how the image processing can be optimized to make informed decisions in cases where the underlying structure is well known from previous experiments. Neural networks, which are trained with simulated data of microtubules, are used to recover the information that is present in any acquired low signal-to-background image. The optimal training parameters are determined and the application on experimental data is presented, outperforming classical algorithms.
Keywords: Microscopy; Two-Photon; GAN