Fluorescence microscopy is an essential tool for investigating functions and internal details of living biological cells. It involves at least three elements: a fluorescent dye, an optical magnifying system (microscope) and an image sensor (camera). This doctoral project is using classical wide-field microscopy combined with CMOS imaging technology for analyzing tightly packed bio-markers (e.g. Toll-Like receptors on lipid rafts) labeled with fluorophores. A problem of interest for this research is that distances between clustered bio-markers are often smaller than the microscope optical resolution (10nm << 250nm).
Current work effort is focused on estimating the amount of fluorophores within imaged cells while dealing with the aforementioned system limitations through image processing techniques. In fine, the methods are extracting metrics to be linked with biological phenomena (e.g. immune cell response). From the gathered images, measurements such as the amount of fluorescent pixels, their intensities or their coarse locations are measured. Correlation between those metrics and the amount of fluorofores are being established and modeled. The developed parametric models offer means to recover biological information directly from image processing results.
The current trends in quantitative fluorescent bio-imaging revolve around super-resolution imaging. Through fitting methods, single-molecule imaging techniques are able to locate precisely a single fluorophore in space. Combined with super-resolution microscopes and by multiplexing the photons emitted by the fluorophores, resolving the location of multiple clustered fluorophores is possible. However, this precision requires costly equipment, careful samples preparation and non-negligible processing power. Compared to the state-of-the-art, the current research aims at developing techniques that also extract locations and quantities of fluorophores, but trading-off the precision to the benefit of lower-end, cost-effective, daily usage lab equipment (e.g. wide-field microscope and non-cooled CMOS imager).
Following the development of image processing techniques, the work done within the frame of this thesis is also focused on hardware implementation. The hardware implementation of the processing methods is motivated by three points. First, stochastic methods need to be applied to mitigate the loss of precision due to the optical setup, which means many data need to be gathered to gain trust in the results. Second, image processing is a good candidate for parallelization, specific algorithms can be very efficiently implemented on reconfigurable hardware (e.g. Field-Programmable Gate Array). And finally, embedding efficient automated signal processing helps biologists gather real-time data on their experiment.
This research is funded by the NutriChip project with a grant from the Swiss Nano-Tera initiative.