Stochastic analysis of lateral resolution and signal-to-noise ratio in fluorescence microscopy : application to structured illumination microscopy

This thesis investigates ways for assessing the practical lateral resolution performance of imaging systems and explores the trade-off between lateral resolution and signal-to-noise ratio (SNR) in high resolution systems. Several authors have commented on the seemingly inherent inefficiency in which imaging systems use the available light when providing resolution enhancement beyond the Abbe limit, often due to the precise mechanism that enabled the resolution enhancement in the first place. Two methods for assessing the lateral resolution and noise performance of microscope systems were developed in this project: a probabilistic analysis based on two-point resolution which has the novelty of being able to deal with synthetic images and the stochastic transfer function (STF) which looks into the effect of noise in the Fourier domain. Results can be determined semi-analytically or by using Monte- Carlo simulations. These methods were applied to several microscope systems such as conventional widefield fluorescence microscopy (WFM) and structured illumination microscopy (SIM) and used to compare their noise performance. These techniques were also used to compare several post-reconstruction processing algorithms for SIM, showing the strengths and weaknesses of each strategy. SIM encodes high spatial frequency information into a senes of images and uses a reconstruction algorithm to yield a high resolution unage. Without further processing, the STF showed that SIM generates noise that is three times greater than that associated with a conventional fluorescence microscope and based on the two-point resolution analysis, the SIM SNR performance in 2D was only comparable with WFM, even though SIM has twice the theoretical bandwidth. However, the reconstruction processing of SIM introduces many redundancies in the form of overlapping spatial frequency orders which can be exploited to improve the SNR further. The STF showed a transition spatial frequency below which the WFM system outperforms SIM and it was shown that a simple WFM-SIM hybrid algorithm based on this observation can indeed significantly improve the SIM result. Common strategies which provide more complete treatments include the Wiener filter and the weighted- average approach. Based on the assumption of uncorrelated noise and a pre-defined goal transfer function, it was shown that the weighted-average method gives the minimum resultant variance statistically. For SIM, this condition is only met when the maximum fringe spatial frequency allowed by the illumination optics is used and the level of noise correlation increases with decreasing illumination grating frequency. The development of a general standardised metric which includes noise considerations can allow more realistic performance assessments of imaging systems and facilitate comparisons. Along with other considerations, such analysis can help practitioners select the most appropriate system for the intended tasks. Understanding the strength and weakness of existing systems will reveal areas of possible enhancements and also help with the development of new techniques.EThOS - Electronic Theses Online ServiceGBUnited Kingdo