MKID Microscopes

Abstract

This thesis describes an investigation of the potential of Microwave Kinetic Inductance Detectors (MKIDs) to be employed as the photon detector in a confocal, fluorescence microscope. As far as can be ascertained, the results presented here are the first ex-vivo observations by such an instrument, an application which has not yet been described in the published literature.
MKID arrays are capable of measuring both the energy and the arrival time of individual photons without false counts, across a wide wavelength range. Consequently they have the potential to be- come an important technology for the next generation of biomedical photonics. These detectors will enable highly-optically efficient, photon-limited, hyperspectral microscopy. This application will be an advantage in a field which benefits considerably from ultra-high sensitivity of photon-limited ima- ging for very low power illumination. Broad wavelength detectors with energy resolution in the near- infrared (NIR) range are of value to the field of bioimaging as more and more instruments are designed to observe longer wavelengths because they allow deeper tissue imaging, with reduced phototoxicity. A significant finding central to this work is the confirmation that MKIDs with a spectral resolution (R) ≈ 10 are able to resolve and separate the spectra of two fluorescent dyes that have been excited by the same illumination source, while also quantifying the number of photons required to attain the re- quired confidence level. Several analytical tools were created and compared using the mean squared error (MSE) method as the comparison metric to determine which was most appropriate to leverage in a given situation based on their simulated performance. These analysis techniques were then applied to three scenarios.
Binomial tests such as the KS test, were found to be of limited use, but did succeed in correctly identi- fying 95.85% of 1000 fluorophores when only given 4 photons each time (assuming all photons were from Nile Red fluorophore or Yellow-Green fluorophore). The number of photons required for a given certainty is highly dependent on, and will increase dramatically if, the spectra observed are closer in wavelength, have a broader spectrum, or if there is a noise source such as autofluorescence. This relationship, while promising is likely to be far higher for fluorophores which are not as widely spectrally separated as the fluorophores chosen, and requires further scrutiny if this is found to be a desirable use case by scientists in the field.
If there is prior knowledge of the fluorescent spectra expected (the reference spectra), then it is possible to employ a mathematical model of the fluorophores, and a least mean squares fitting technique, such as χ2 analysis or scipy.curve_fit. It appeared to be the case that scipy.curve_fit provided a better fit in low photon counts, whereas χ2 was superior in high photon counts but this was not thoroughly tested. The absence of readout noise meant subjects could be positively identified to a confidence of 5σ using as few as between 43 and 84 photons.
In the absence of fluorophores (reference spectra), two blind spectral un-mixing methods were attemp- ted. Non-negative matrix factorisation (NMF) was found in analysis to have a similar effectiveness to non-blind techniques (with known fluorophores reference spectra). This blind technique was applied to the ex vivo sample of a nematode worm (see Chapter 8.3) to successfully obtain an image which resembled a reference image taken from a commercial microscope, confirming the MKID microscope was indeed capable of discerning between fluorescent markers based on spectral information.
This research found that MKIDs are a powerful and effective tool, one that promises to be particularly advantageous in the field of biomedical photonics. These developments provide a sound foundation for further work in this domain and related fields.