Information optimised sensor-less adaptive optics for multiphoton fluorescence microscopy

Thesis

Multiphoton fluorescence microscopy is an important technique to image biological structures especially through deep tissue. The microscope images can be degraded by aberrations, so adaptive optics (AO) is normally required to correct aberrations using adaptive elements introduced into the microscope. In many cases, AO is operated without a wavefront sensor to directly measure the aberration but with images taken sequentially for different pre-defined modulations called biases to indirectly estimate the aberration. The sensor-less AO allows a simple optical design thus favoured in a wide range of applications, but it tends to be time-consuming due to the sequential measurement of multiple images and the need for iterations to reach satisfactory estimation. For more time-efficient performance, sensor-less AO methods have been developed to estimate the aberration in more suitable representations and by more powerful algorithms. The state-of-the-art performance has been achieved with a deep learning (DL) based algorithm, which can estimate the aberration from at least two images. However, these methods generally operate image measurement in a heuristic way hence the images are not necessarily in the best condition for aberration estimation. This has prevented sensor-less AO achieving its optimal performance.

In this thesis, sensor-less AO was optimised by considering the information contained in the images. First, a framework was introduced to analyse the information in the images by Fisher information and derive a loss function based on Cramér-Rao bound to optimise the biases; this framework is not specific to any sensor-less AO method or imaging scenario; the results showed that the optimised biases brought better sensor-less AO performance and the loss function derived from information analysis served a good performance estimator. Then, a new sensor-less AO method was proposed to parallelise the sequential image measurement by using an array of focal spots for imaging with each spot containing a different bias optimised by the framework introduced above, so that the information used to be separated in multiple images was concentrated into one single superimposed image; with a DL based algorithm to estimate the aberration from the superimposed image, the method could finish in milliseconds with even a single image measurement; experiments on a two-photon (2-P) microscope showed that the method generally maintained effective performance in a wide range of scenarios. This optimised sensor-less AO has great potential to improve the imaging performance of multiphoton fluorescence microscopy, especially in challenging situations with fast-changing aberrations. The benefits may also be extended to other imaging modalities.

Authors: Zhang, B

• DOI: 10.5287/ora-xqaxryqb7
• Type of award: DPhil
• Level of award: Doctoral
• Awarding institution: University of Oxford
• Language: English
• Keywords: sensorless adaptive optics; information optimisation; fluorescence microscopy
• Subjects: Optics, Adaptive