In vivo autofluorescence and specific fluorescence unmixing by Non negative Matrix Factorization

International audienceFluorescence imaging locates fluorescent markers that specifically bind to tumors: markers are injected to a patient, and optimally excited with near infrared light. To investigate thick media, as the fluorescence signal decreases with the light travel distance, the autofluorescence of biological tissues becomes a limiting factor. To remove autofluorescence and isolate specific fluorescence, a spectroscopic approach, based on Non-negative Matrix Factorization (NMF), is explored. A murine breast cancer cell line (5.106 TS/Apc cells in 200 µL sterile DPBS) was implanted subcutaneously in female nude mouse. Twelve days after tumor cells implantation, fluorescent markers (AngioStamp,Fluoptics-Grenoble-France) were injected intravenously (50 nmol, 200 µL). Two days after marker injection, a spectral acquisition is performed. The animal is illuminated with a laser at 690 nm and the emitted back fluorescence signal is collected by an imaging spectrometer coupled with a CCD camera. At this point, the ratio between non-specific fluorescence (marker everywhere but in the tumor) and specific fluorescence (marker in the tumor) is supposed to be optimal for biological analysis, but a non-negligible autofluorescence signal (with a distinct fluorescence spectrum) still remains. We process our NMF algorithm on the spectrally resolved acquisition: the algorithm successfully separated fluorescent markers from autofluorescence despite their overlapping fluorescence spectra. Unmixing results are presented on figure 1: non-specific fluorescent markers signal still remains in kidneys and everywhere else the marker has traveled, but autofluorescence has been removed. Ratio between skin and tumor has been noticeably increased. Comparable results have been obtained from two hours after injection, with only 5 nmol (200µL) of marker: tumor localization was already sensibly improved. We found that in vivo spectrally resolved acquisition combined to NMF processing successfully filters fluorescence contributions of interest from measurements impaired by unwanted signals. Autofluorescence removal is a fundamental pre-processing step to get accurate FDOT reconstructions