Fluorescence microscopy combines the power of high performance optical components and digital image analysis. The emission signals of fluorescent dyes show a significant spectral crosstalk which inevitably increases with the number of fluorescent dyes used. However, it is possible to significantly reduce the effects of spectral crosstalk by the application of spectral imaging and spectral unmixing techniques. Spectral imaging is the combination of computer vision and spectroscopy.

Epidemiology of Breast Cancer

Breast cancer is caused by a malfunction in the cellular mechanisms which regulate cell growth [3]. Mutations, such as point mutation, chromosome translocation and gene-amplification, can cause these proto-oncogenes to change their behavior and become hyperactive and even non-physiological. Proto-oncogenes are genes that are responsible for the development and differentiation of cells. HER-2/neu is a proto-oncogene and one of a few evidence-based features for the diagnosis of breast carcinoma [3]. Normal breast epithelial cells have two HER-2/neu gene copies and between 20000 and 40000 HER-2/neu receptors. In the early stages of 20% of breast carcinomas the HER-2/neu is over expressed because of gene-amplification [4].

Fluorescence In-Situ Hybridization (FISH)

FISH is a cytogenetic technique that is used to detect and localize the presence or absence of specific DNA sequences on chromosomes. It uses fluorescent dyes that bind to only those parts of the chromosome which show a high degree of sequence similarity. Fluorescence microscopy can be used to analyze the fluorescently labeled tissue preparations. An example for a FISH analysis of human tissue can be seen in the RGB image in Fig. 1(d). The sample was stained with three different fluorescent dyes: (i) DAPI (4',6-diamidino-2-phenylindole) to stain the cell nuclei, (ii) SpectrumOrange and, (iii) FITC (fluorescein-5-isothiocyanate) to stain two specific parts inside the cell nuclei (HER-2/neu and CEP 17). In practice, a skilled pathologist would manually count the red (SpectrumOrange) and green (FITC) signals inside every blue (DAPI) cell nucleus.

With three band-pass filters, one for each fluorescent dye, it is possible to separate the emission signals of the fluorescent dyes. But due to the spectral crosstalk of the fluorochromes the emission signals overlap to a certain degree. The resulting fluorescent images of the green, red and blue fluorescent dye can be seen in Fig.1 (a, b, c) respectively. The effect of the spectral crosstalk between the emission spectra of the three fluorescent dyes is visible in the red and the green channels where the shape of the nuclei is slightly visible, as an artifact. This makes the segmentation of the red and green signals a challenging task. Since the spectral crosstalk inevitably increases with the number of dyes used, hyper-spectral imaging can provide better results because of the higher spectral resolution [7].

Fluorescence Microscopy

Fluorescence has been a useful tool for a wide variety of biological applications since ages. The fluorescence effect occurs when a molecule is excited and emits a photon as it relaxes to a ground state. Some materials exhibit fluorescence naturally (auto fluorescence) while others require a dye or a marker. Different molecules or probes will absorb and emit different wavelength ranges. Many of the commonly used fluorescence markers are excited with UV wavelengths and emit UV or visible wavelengths. In many cases fluorescence is paired with microscopy to observe the emission signals of the excited molecules. Most fluorescence microscopy systems follow the same principle: the sample is exposed to an excitation source and the fluorescently labeled parts of the sample emit a fluorescence, which is collected. The most basic setup for fluorescence system has an excitation channel and an emission channel separated by a dichroic beamsplitter. The excitation channel has an excitation source and excitation filter. The excitation filter transmits only the desired excitation wavelength. The emission channel has an emission filter, which transmits only the emission band of the fluorochromes and blocks the excitation wavelength. A dichroic beamsplitter is used between the two channels to ensure the appropriate amount of exposure with the excitation source and maximum transmission of the emission wavelengths. The objective lens focuses the excitation source onto the sample and collects the emission signal [7].

Spectral Imaging (SI)

There are different methodologies to acquire hyper-spectral data. The classical approach is to spatially scan a sample with a single point probe while recording spectral data for each point.
This approach provides both spectral and spatial resolution, but, due to the acquisition time is not applicable for real-time applications. For the acquisition of a three dimensional hyper-spectral data cube shown in Fig. 3, either the wavelength information or one spatial dimension must be acquired sequentially. For this paper a wavelength scanning SI system utilizing a liquid crystal tunable filter was used. Wavelength scanning SI is essentially based on acquiring a number of monochrome images of a sample at different wavelengths. Both spatial dimensions are acquired simultaneously, while the spectral information is acquired sequentially. After completion of the data acquisition, the single, spectrally encoded images are combined, allowing the calculation of the spectra for each pixel. For hyper-spectral images, the acquisition of a single, spectrally highly resolved spectroscopic image may last some tens of seconds. Thus this method is useful particularly when only a few images at characteristic wavelengths have to be recorded, e.g. for fluorescence microscopy. Chemometric evaluation algorithms can only be applied after having recorded the full spectral hypercube (see Fig. 3). This necessarily involves handling of large amounts of data, which makes demands on the storage and processing capabilities of the computer system [7].