Spectral Unmixing Technique For Fluorescence Reflectance Imaging

Fluorescence reflectance imaging is a powerful technique for detecting labeled cellsand/or molecules in-vivo and tracking the formation and development of disease, due to itsrelatively continuous, non-invasive and high-throughput advantages. However, theshortcoming of this imaging technique lying in the presence of autofluorescence fromfood and skin will greatly limit sensitivity and make it difficult to accurately monitor andlocate the fluorophores of interest. In addition, in order to monitor a variety of biologicalprocesses simultaneously, it is necessary to use multiple fluorescent markers to labeldifferent molecules. Multi-spectral unmixing can remove autofluorescence and separatemultiple fluorophores of interest.In this thesis, we proposed a multi-spectral unmixing method: firstly, extract the purespectral of each component from5-6pieces of multi-spectral fluorescent images, thenlinear unmixing algorithm was used to remove autofluorescence and separate differenttarget fluorophores. This algorithm was applied to the fluorescence reflectance imagingsystem, which removed autofluorescence and achieved the unmixing of E.coli-TagRFPand E.coli-mLumin. The signal to noise ratios between these two fluorophores andautofluorescence increased from9.23dB and4.70dB to35.69dB and24.91dB, respectively.In addition, in the case of faint fluorophores of interest, where it is difficult to predictspatial distribution of target fluorophores, we need to know their positions in advancebefore extracting their pure spectral for linear unmixing. Thus, we improved this algorithm.A classification algorithm after refining initial points was used to obtain the spatialdistribution of target fluorophores. In-vivo model experiment and in-vivo nasopharyngealcancer tumor experiment further verified the improved linear unmixing algorithm.