Fluorescence imaging is an indispensable tool for biological studies, especially in cellular research. Through the staining of cells with various fluorophores and imaging under a microscope, a large volume of color-coded processes can be quantitatively characterized, from chromosome dynamics [1
] to gene expression [2
]. The development of fluorescent probes has greatly improved cellular research by bringing high quantum yield fluorescent dyes and multiplexed staining methods into the field [3
]. This has enabled researchers to investigate several organelles and their interactions in the same field of view (FOV) and at the same time with high contrast. As more and more fluorophores are developed in the visible light band, a central problem for multi-staining methods to solve is how to discriminate fluorescent probes with spectral peaks very close to each other. So in addition to the traditional requirements on spatial resolution, high spectral resolution is also necessary for imaging devices which target fluorescent imaging applications.
Hyperspectral fluorescence microscopy (HFM) is an emerging field based on hyperspectral or multispectral imaging concepts, which often borrow from remote sensing techniques [4
]. Because of its high spectral resolution (less than 10nm), HFM has found many applications in spectral imaging of living cells [6
]. HFM has also been used to discriminate the contributions of autofluorescence from exogenous fluorescent signals present in the sample [10
]. Compared to only 3 spectral bands that can be obtained by the traditional RGB color cameras or multi-filter imaging, HFM has the capability to capture the whole fluorescent spectrum within its 2D FOV, and build a 3D datacube (x, y, λ) for multivariate data analysis. Such data can provide accurate information about fluorescent probe distributions and their relative concentration over the whole specimen.
On the other hand, HFM with high temporal
resolution is gaining more importance in biological microscopy. This is because it can be used to capture transient scenes, which is often a critical requirement in cellular dynamics research. Unfortunately, most currently available HFMs need scanning which limits their temporal resolution. For example, hyperspectral confocal microscopy is a spatial scanning technique which can implement three-dimensional sectioning while providing spectral information. However, even state-of-art HFM/confocal systems can only acquire data at a rate up to 5 frames/s with 512×512 pixels [11
]. HFM with an acousto-optic tunable filter (AOTF) or liquid crystal tunable filter (LCTF) is another technique based on spectral scanning [12
], which can switch wavelengths very fast. For HFM with AOTF, switching times are typically less than 100 microseconds [14
]; for HFM with LCTF, around 50ms in the visible light band, and ~150ms in NIR band [15
]. However, there is a trade-off between the quantity of spectral bands captured and the total acquisition time. Plus, due to fairly poor throughput (the AOTF has transmission of ~30% in the visible light range [14
]; LCTF can exhibit over 50% peak transmission for red and NIR light, but this number reduces to ~15% in the blue region [15
]), these systems are not ideal candidates for fluorescent real-time imaging. In addition to the above HFM approaches, other scanning techniques include Fourier-transform imaging spectrometers [16
] (scanning in phase space) and fiber Fabry-Perot arrays [17
] (scanning in frequency). The scanning mechanism of these HFMs still decreases their temporal resolution and limits their potential use in real-time imaging. To fully utilize the potential information yielded by fluorescent probes in HFM, snapshot techniques are needed.
Currently, many snapshot techniques have been developed for hyperspectral imaging, such as aperture splitting [18
], field splitting (by fibers [19
] or lenslet arrays [20
]), Computed Tomography Imaging Spectrometry (CTIS) [21
], and Coded Aperture Snapshot Spectral Imaging (CASSI) [22
]. Among these, CTIS and CASSI are particularly interesting due to their higher throughput and compact size which are both critical features for fluorescence microscopy. CTIS has already been demonstrated in fluorescence microscopy [24
] and CASSI has just recently been used in real-time spectral imaging for remote sensing applications [25
] but has yet to be tested in microscopy. CTIS utilizes a computer-generated-hologram (CGH) to map multiple projections of the 3D data tube (x, y, λ) onto a 2D detector array. After being processed by linear algebra reconstruction methods, spectra from every spatial position within the CGH's two-dimensional FOV is collected. Although CTIS can provide spectral imaging of fast moving and/or low-light objects it suffers from many problems, including massive computational requirements and the missing cone effect. CASSI draws on the ideas of compressed sensing. The spatial modulation is brought in by a coded aperture, and is later transformed to spatial and spectral modulation in the undoing process. Then a multiscale reconstruction algorithm is employed to extract the spatial and spectral information from the mask-modulated intensity graph. However, this technique has limitations on spectrally resolving point sources. Beside these two, current aperture splitting and field splitting techniques also have defects. Aperture splitting is not light efficient, while field splitting by fibers or lenslet arrays is limited by size of their spatial sampling components.
In this paper, we present a novel snapshot HFM device - the Imaging Slicing Spectrometer (ISS). It can acquire the whole spectral information within its FOV via a single integration of an array detector. By directly imaging the remapped and dispersed image zones onto a CCD detector, the ISS system overcomes the CTIS and CASSI's problems of computational reconstruction and resolution loss. The ISS acquires data directly with minimal post processing to build a 3D datacube.
Although the ISS concept has already been established in astronomical optics for over a decade [26
], because of the characteristics of imaging objects (galaxies, stars, etc.), astronomical ISS systems have relatively low spatial sampling (typically less than 60 image slices) [30
]. No current astronomical ISS system can be simply modified and adapted for the demanding requirements of biological fluorescence microscopy. To the best of our knowledge, this is the first time that the image slicing concept has been implemented for high-resolution microscopy. Our ISS system can be easily coupled to any microscope system with an image output port. This prototype surpasses existing astronomical ISS systems in two aspects. First, the slicing component has been miniaturized. The width of the slicing component in the image slicer is quite small (160μm) compared to the FOV (25mm), enabling high spatial sampling of the object. Second, use of a single large-format CCD detector combined with grouped 2D slicing directions greatly simplifies the reimaging process. This is an important improvement since it allows the construction of a compact high-resolution system. The prototype realizes 100×100×25 sampling in the 3D datacube (x, y, λ), which corresponds to 0.45 microns and 5.6 nm resolution in spatial and spectral domains respectively. While the system presented here is a proof-of-concept device, if required, the instrument could be redesigned and built to a different specification to improve spatial and spectral resolution. The imaging results presented in Section 4 demonstrate the promising potential of the ISS system in HFM research.