Image analysis of lipid deposits in living organisms visualized by CARS microscopy

CARS microscopy is a novel technique for imaging bio-molecules in living cells using the molecular vibration as contrast mechanism. Global and local thresholding, state-of-the art image analysis techniques (watersheds, level sets) and a recent technique for anisotropic Gaussian fitting are used to segment lipid droplets in CARS microscopy images of, S. cerevisiae (yeast cells). The ability to extract quantitative information, such as vesicle size, is validated by means of images of polystyrene beads of well-defined size. Theoretical modeling of the CARS signal is performed, linking the physical object to its representation as a CARS microscopy image, as an attempt to take the non-linear relationship into account for refined image analysis. Theoretically modeled images of polystyrene beads correspond well to those experimentally obtained. Of the image analysis tools, it is found that global and local thresholding have limited use for segmentation of CARS microscopy images. Watershed and level set segmentation underestimate the size of a segmented object. Anisotropic Gaussian fitting overestimates the size of the object in the presence of non-resonant background. Reduced overestimation is achieved by background suppression. This emphasizes the importance of background removal in the CARS microscopy images, both experimentally as well as analytically. Automated image analysis of CARS microscopy images of S. cerevisiae is shown to be an excellent approach for in-depth studies of fat deposits in living organisms, with prospects for including human cells. Such a methodology will be of outmost importance for improved understanding of the mechanisms behind the development of obesity.