Trilinear analysis of images obtained with a hyperspectral imaging confocal microscope

Hyperspectral imaging confocal microscopy (HSI-CM) is a powerful tool for the analysis of cellular processes such as the immune response. HSI-CM is a data rich technique that routinely generates two-way data having a spectral domain and an image or concentration domain. Using a variety of modifications to the instrument or experimental protocols, one can readily produce three-way data with HSI-CM. These data are often amenable to trilinear analysis. For example we have used a time series of 18 images acquired during photobleaching of the fluorophores in an effort to identify fluorescence resonance energy transfer (FRET). The resulting images represent intensity as a function of concentration, wavelength and photodegradation in time, to which we apply our techniques of trilinear decomposition. We have successfully employed trilinear decomposition of photobleaching spectral image data from fixed A549 cells transfected with yellow and green fluorescent proteins (YFP and GFP) as molecular probes of cellular proteins involved in the cellular immune response. While useful in the interpretation biological processes, the size of the data generated with the HSI-CM can be difficult to manage computationally. The 208 x 204 x 512 x 18 elements in the image data require careful processing and efficient analysis algorithms. Accordingly, we have implemented fast algorithms that can quickly perform the trilinear decomposition. In this paper we describe how three-way data are produced and the methods we have used to process them. Specifically, we show that co-adding spectra in a spatial neighborhood is a highly effective method for improving the performance of these algorithms without sacrificing resolution.