Blind Decomposition of Multi-spectral Fluorescence Lifetime Imaging Microscopy Data: Further Validation

摘要

Characterization of living tissue without the need for biopsies is the goal of several probe technologies such as Multi-spectral Fluorescence Lifetime Imaging Microscopy. This technique measures the mixed response from the endogenous fluorophores within an organic sample. This response is decomposed into the individual response from every constituent using a fully constrained linear unmixing algorithm: Blind End-member and Abundance Extraction (BEAE). Further validation of the method is needed specially when dealing with real laboratory samples. Moreover, the BEAE method incorporates a regularization parameter during the quadratic optimization procedure which has to be tuned to improve the estimation accuracy. Different values for the regularization parameter are tested using synthetic data at a signal-to-noise ratio of 10 dB and 15 dB. The relative error against the ideal end-members for each component is measured. Results show that the estimation accuracy in each end-member increases when the regularization parameter is around 0.75. Blind decomposition of m-FLIM data from coronary samples is also performed for validation purposes. The extracted fluorescence decays are identified as collagen, elastin and LDL responses. Histopatology slides are used as reference to validate the results. Synthetic simulation shows that the BEAE algorithm performs a more accurate estimation of the end-members profiles due to the regularization term. Furthermore, analysis performed on ex-vivo samples match the qualitative description provided by histopatology slides.