3.1.

Whiskbroom

The whiskbroom imaging mode, also known as the point-scanning method, was used originally by the Earth Resources Technology Satellites and then implemented by the Airborne Visible/Infrared Imaging Spectrometer. In the whiskbroom sensor, rotating mirrors were often used to scan the landscape from side to side perpendicular to the direction of the sensor platform. The rotating mirrors redirect the reflected light to a point where a single or just a few sensor detectors are grouped together. As shown in Fig. 2(a), a single point is scanned along two spatial dimensions ($x$ and $y$) by moving either the sample or the detector. Then the reflected light is dispersed by a prism and recorded by a linear array detector. The spectral image data cubes ($x$, $y$, $\lambda$) can be acquired by scanning over the 2-D scene ($x$, $y$) and by dispersing in the wavelength domain ($\lambda$). In the whiskbroom mode, two-axis motorized positioning tables are usually needed to finish the image acquisition, which makes the hardware configuration complex. In addition, this kind of imaging mode is also usually time-consuming because it needs to scan both in the $x$ and $y$ spatial dimensions. Therefore, another scanning approach with high acquisition speed and high sensitivity has been proposed, that is, the spectral confocal laser-scanning method. This method is usually accomplished by illuminating and acquiring images through confocal or conjugate pinholes to restrict the in-focus optical section thickness and by incorporating multiple lasers and wavelength dispersive spectrophotometers to get spectral information. An advantage afforded by this method is the ability to control depth of field, eliminate or reduce background information away from the focal plane, and the capability to collect serial optical sections from thick specimens. Another advantage of the whiskbroom imaging is that it has fewer sensor detectors to keep calibrated as compared with other types of sensors. Most commercial spectral confocal scanning instruments belong to this kind of imaging mode.

Fig. 2

Typical spectral imaging approaches. (a) Whiskbroom. (b) Pushbroom. (c) Staring. (d) Snapshot.

3.2.

Pushbroom

The pushbroom approach, also known as line scanning, has been used by some earth observation satellite systems operating from space, such as the SPOT system and the Advanced Land Imager (ALI). Different from the whiskbroom method that scans one point at a time, the pushbroom method can simultaneously acquire a slit of spatial information, as well as spectral information corresponding to each spatial point in the slit for one scan, that is, a special $y-\lambda$ image with one spatial dimension ($y$) and one spectral dimension ($\lambda$) can be taken at the same time with a matrix detector. As shown in Fig. 2(b), the pushbroom method collects a slit image from the object dispersed onto a 2-D detector, in which the spatial information is displayed along one axis and wavelength information along the other. Then the spectral image data cube can be obtained by scanning the slit in the direction of another spatial axis. A pushbroom scanner can get more light than the whiskbroom one, since it can stay at a particular area for a longer time providing a long exposure on the matrix detector and a relatively high spectral resolution as well.⁸ However, in most of the pushbroom imagers, either the camera or the object should move accordingly as one line of the image is acquired at each scan and the move should be synchronized with the frame acquisition rate of the array detector in order to ensure the measurement of a smooth image. Another approach to perform the pushbroom procedure is to put a movable slit aperture or a rolling shutter into the conjugate image plane to eliminate out-of-focus light. This method allows for higher illumination transmission while exhibiting only a slight increase in the confocal depth response, which has been widely used in multispectral line confocal imaging systems.⁹

3.3.

Staring

The staring approach (also known as band sequential method) is a spectral scanning method that acquires a single band 2-D gray-scale image ($x$, $y$) with full spatial information at once. As shown in Fig. 2(c), this mode generally uses filters [such as filter wheels containing fixed bandpass filters, linear variable filters (LVF or wedge filters), variable interference filter (VIF), and tunable filters] instead of a grating or prism in front of a matrix detector. The light passes through the focusing optics. Then it is filtered, which induces only a narrowband segment of the spectrum that impinges on the focal plane of the detector (typically is a matrix CCD). Therefore, the 2-D image in one wavelength can be captured at a given time, and the image cube is filled by tuning the output wavelength of the filter as a function of time.¹⁰ In the staring mode, the entire scene is imaged onto the focal plane, one spectral band at a time, and goes through the spectral wavelengths, which is different from the whiskbroom and pushbroom approaches. Therefore, the number of spectral bands acquired is completely up to the user. Also, and importantly, a dynamic range can be preserved, since different exposures can be employed at different wavelengths.

3.4.

Snapshot

The snapshot (also known as single-shot) method is intended to record both spatial and spectral information on an area detector with one exposure. Unlike the whiskbroom, pushbroom, and staring modes that need to scan either in the spatial dimension or in the spectral dimension, thereby limiting their temporal resolution, the snapshot mode is an imaging technique with no need to scan at all. As shown in Fig. 2(d), the snapshot mode can acquire a complete spectral data cube in a single integration by directly imaging the remapped and dispersed image zones onto a CCD detector simultaneously.¹¹ Although this mode can acquire data directly with minimal postprocessing to build a 3-D data cube, its spatial and spectral resolutions are limited as the total number of voxels cannot exceed the total number of pixels on the CCD camera. Therefore, for a given CCD, one can always increase the spatial sampling at the expense of spectral sampling, and vice versa.¹²

3.5.

Comparison of Different Approaches

Among these four kinds of spectral imaging approaches, each one had both advantages and disadvantages when they were applied to analyze biological tissues. Table 2 shows the comparison of the whiskbroom, pushbroom, staring, and snapshot imaging modes. There is no “absolutely” best mode, but in order to select the most suitable one for a given biomedical application, the characters of different biological tissues and the detection objective must be considered.

Table 2

Comparisons of whiskbroom, pushbroom, staring, and snapshot approaches.

From Table 2, it can be seen that the whiskbroom and pushbroom modes use a dispersive element [prism, grating, prism-grating-prism (PGP), etc.] to split the light, a factor that gives an advantage of uniformly high efficiency, low scatter, and low cost to these kinds of imagers. Another advantage of the pushbroom mode is that the collected images reflect only the $y-\lambda$ plane, making a single line of light feasible as the source for image illumination.¹³ This approach is suitable for applications of searching an object with specific spectral signatures as the entire spectrum of each pixel in the slit image is available in real time. In addition, systems that are based on dispersion elements, such as prisms and gratings, are also efficient for measuring the spectral images through confocal microscopes, which implement the scanning procedure by scanning optics.¹⁴ The whiskbroom and pushbroom modes also have disadvantages. The optical design is complex and the data cube collection time is long for conventional mechanical scanning systems. Although some optical scanning systems such as the Zeiss and Nikon laser scanning systems can capture data as fast as the snapshot systems, the acquisition of high-contrast images commonly accompanies strong photobleaching, which is of particular concern in prolonged imaging experiments.¹⁵ In general, the whiskbroom and pushbroom imagers were often coupled with microscopes and used for fluorescent and histopathological analyses.

The most important advantages of the staring mode are the short collection time of data cube and the ease of coupling with other medical optical instruments, as no relative spatial move is required between the sample and the detector. Another advantage is the flexibility of selecting the optimal spectral range for each application. These advantages make it suitable for a wide range of biomedical applications, such as detecting organs, tissues, and cells in vivo or in vitro when it is coupled with a camera lens, endoscope, or microscope.^13,16 There are also some disadvantages of using this mode, such as a lower spectral resolution, lower transmission rate (typically reaches only 35 to 60% depending on the type of device and model), and a higher cost than the whiskbroom and pushbroom modes (except the laser scanning systems). Although tunable filters attempt to shorten the total capture time by selectively acquiring spectral bands, or by altering each channel’s exposure time as a function of wavelength so that the total data capture time can be saved in bright channels, there are still shortcomings that can be difficult to overcome, such as the low throughput and sequential acquisition mode, particularly in multifluorophore imaging, which requires at least 30 spectral channels to be captured.¹⁵ The interferometer-based spectral imaging systems, however, usually have high optical throughput, high and variable spectral resolution, mechanical/temperature stability, and flexible selection of wavelength range. They still have limitations for practical applications, such as the varying sensitivity in the entire spectral region and the computation intensity for data transform.¹⁷

For the snapshot mode, no scanning in either spatial or spectral domains is needed for obtaining a spectral data cube, making it attractive for applications requiring fast spectral image acquisitions. Therefore, it is an ideal solution for applications where imaging time is a major factor. It is suitable for time-sensitive imaging experiments, such as the observation of fast-diffusing molecules or the determination of temporally resolved dynamic biological processes. It is also suitable for detecting oxygen saturation maps in the retina of human eyes when it is coupled with a fundus camera. This has been demonstrated by Johnson et al. with computed tomography imaging spectrometer (CTIS) with an exposure time of $\sim 3\mathrm{ms}$.¹⁸ However, this mode is still in the early stages and not fully developed. Only a few implementations that rely on complicated fore-optics design and computationally intensive postprocessing for image reconstructions are currently available, with limitations for range and resolution of spatial and spectral dimensions.¹⁷

In summary, each spectral imaging mode has advantages and disadvantages when used in biomedical analysis. For example, some whiskbroom and pushbroom systems can get high spectral and spatial resolution and are low costing, but their hardware assembly is usually complex. The staring mode can be assembled compactly and coupled easily with other optical instruments, but it is more expensive than the others. The snapshot mode can quickly capture spectral data cube, but its spectral and spatial resolutions are limited. However, the advantages of different modes can be exploited fully if they are selected correctly and used according to the objectives of different biomedical analyses. We will discuss this in detail in the following sections.

4.1.

System Architecture

To acquire a 3-D data cube, spatial (2-D) and wavelength as a third dimension of a biological tissue, a spectral imaging system suitable for biomedical applications needs to be developed. As shown in Fig. 3, the hardware architecture of a typical biomedical spectral imaging system generally consists of four parts: collection optics or instruments, spectral dispersion element, detector, and system control and data collection module.

Fig. 3

The system architecture of a typical biomedical spectral imaging system. The reflected or transmitted light collected by the collection optics passes through the spectral dispersion element and is focused on the detector. The system control and data collection module can control the detector and the spectral dispersion element, capture the spectral images, and output results.

The collection optics or instruments part refers to the collection and imaging optics such as microscopes, endoscopes, camera lens, etc., which can produce an image on the spectral dispersion element. The spectral dispersion element is the heart of the system that enables the separation of the light into different wavelengths. Dispersive spectrometers and spectral filters are two kinds of commonly used wavelength dispersion devices. The dispersive spectrometer is a collection of transmitting or reflecting elements separated by a distance comparable to the wavelength of the light under investigation. It often refers to prisms, gratings, PGP, or beam splitters, which can be used with whiskbroom, pushbroom, and some snapshot imaging modes. Spectral filters have the characteristic to pass radiation through a very narrow bandpass, or spectral bin, and this can be spectrally tuned over a wide spectral range, usually in a very short time. Acousto-optical tunable filter (AOTF), liquid crystal tunable filter (LCTF), and circular and linear variable filters (CVF and LVF) are the most popularly used tunable filters in spectral scanning instruments. Another generally used spectral dispersion element is the interferometer, which can obtain spectral images by changing optical path-length difference (OPD) between two beams, recording the interference intensity as a function of the OPD (interferogram), and Fourier transforming the interferograms pixel by pixel.¹⁹ For example, the Michelson interferometer and the Sagnac interferometer have been used to develop some spectral imaging systems for different biomedical research.^20,21 The detector is used to acquire the light intensity required to measure the spectrum at each pixel in the image. Some sensors such as the photomultiplier tube, linear CCD, area CCD and CMOS, focal plane array, and electron multiplying charge coupled device (EMCCD) can be used as a detector according to different imaging requirements. In recent years, the scientific CMOS (sCMOS) has been introduced and also used in biomedical imaging as it offers an advanced set of performance features, such as high resolution, without compromising read noise, dynamic range, or frame rate. These features make the sCMOS ideal for high-fidelity, quantitative scientific measurement. Therefore, some commercial sCMOS [for example, the ORCA-Flash2.8 and Flash4.0 camera (Hamamatsu), Andor and PCO series sensors, and Rolera™ Bolt Camera (QImaging)] can also be used as a detector of spectral imaging system.

In the past decades, researchers have developed various spectral imaging systems for biochemical analysis on various biological organs, tissues, and cells. Table 3 shows some selected spectral imaging systems used in the biomedical field. We will break up the discussion of instrument configuration according to the imaging mode in the following subsections.

Table 3

Current biomedical spectral imaging systems.

4.2.

Current Biomedical Spectral Imaging Systems

4.3.

System Performance

It is important to evaluate the performance of a biomedical spectral imaging system since different applications usually require different performance parameters. The main performance parameters of a biomedical spectral imaging system generally include spatial resolution, spectral range, spectral resolution, dynamic range, sensitivity, etc. These parameters depend mainly on the capability of the optical system (the optical head, the spectral dispersion element) to produce an undistorted, unaberrated, and intense data cube of a scene, as well as the ability of the detector to convert an image into a digital form with low noise and fine spatial or spectral sampling.¹⁶¹ Some performance parameters of current biomedical spectral imaging systems have been listed in Table 3. A brief discussion of the most important performance parameters and their relationship with the applications follows.

4.3.1.

Spatial resolution

In biomedical imaging, the spatial structure information is one of the important factors for histological analysis and disease diagnosis. The inclusion of spatial information can increase the biological information content derived from data acquired with spectral imaging systems. As shown in Fig. 4, different biological objects usually have different spatial scale requirements for detection. Spatial resolution of a spectral imaging system determines the closest distinguishable features in the objects. With a proper selection of the system configuration, useful spatial information and quantification can be achieved by sufficient biological sampling, as different spectral imaging methods usually have different spatial resolutions ranging from a few micrometers to a few centimeters. For example, cells expressing each color of fluorescent protein can be located and clusters of cells expressing two or three colors usually reinforce the need for increased spatial resolution ($<1\mu \text{m}$) to facilitate single cell detection.²⁶ A spectral imaging system with spatial resolution of 0.45 mm and 17-cm-diameter field of view (FOV) is sufficient for hemoglobin oxygen saturation detection in sickle cell disease patients and in some other clinical investigations.^70,73 Therefore, proper selection of spatial resolution is important for different kinds of biomedical applications.

Fig. 4

Schematic representation of spatial scale of biological objects. As different biological objects generally have different spatial scales ranging from a few micrometers to a few centimeters, a proper selection of spatial resolution is usually needed to detect the object properly.

Spatial resolution defines the level of spatial detail depicted in an image, which determines the closest distinguishable features in an object. The theoretical spatial resolution mainly depends on two factors: the wavelength ($\lambda$) and the numerical aperture (NA) of optics. In addition, the pixel size of the detector and the magnification also play a role because they determine the sampling frequency, which must be sufficiently high to achieve full resolution.¹⁶² The spatial resolution of most spectral imaging systems can dominantly deteriorate by astigmatism, which is an off-axis aberration that increases with the square of both the off-axis angle and NA.¹⁶³ Significant loss in spatial resolution due to prism mount misalignment in a chromotomosynthetic HSI system has been analyzed by Bostick et al. They found that the relative effect is less in the shorter-wavelength region, which indicates that spatial performance is also affected by wavelength.¹⁴⁹ However, astigmatism is not a problem in a spectral confocal laser scanning system because the laser submits single points from the FOV sequentially, not multiple points simultaneously. For a line-scanning spectral imaging system, the spatial resolution in the direction of the slit width is affected by the slit width, whereas the spatial resolution along the slit length is usually affected by the magnification and camera pixel size,⁴⁴ while for a spectral filters-based imaging system, the spatial resolution is also related to the types of tunable filters (AOTF, LCTF, etc.) the system used.^4,70,87 Currently, spectral images acquired by microscopic biomedical spectral imaging systems are usually in the order of 0.3 μm or finer, whereas those systems coupled with camera lens generally acquire higher spatial resolution data, usually in the order of 0.45 mm or finer.^44,70

There is also another problem to consider in the bright-field light microscopy and fluorescence microscopy combined with spectral imaging function. That is, how to produce high spatial resolution or clear images of a focal plane deep within a thick tissue. For example, a fluorescent cell might be 5 to 15 μm thick, whereas the thickness of the imaging plane of a high-NA objective ($\mathrm{NA}\ge 1.3$) is only $\sim 300\mathrm{nm}$ or less. Thus, the vast majority of the cell volume is out of focus and becomes blurred.¹⁶⁴ Optical sectioning may be one of the approaches used in solving this problem. It can get better background rejection, improved contrast, and more accurate spectral characterization by reducing out-of-focus artifacts. Compared to common imaging with wide field, optical sectioning generally led to greater than $3\times$ improvement in the rejection of out-of-focus fluorescence emissions and $\sim 6\times$ greater peak-to-background ratios in biological specimens, yielding better contrast and spectral characterization, which is important in unmixing signals in biological samples and suppressing a phototoxicity and bleaching of fluorophores.¹⁶⁵ The commonly used methods for optical sectioning are confocal laser scanning microscopy and two-photon microscopy. These methods are powerful means to eliminate the background caused by out-of-focus light and scatter from images and are well matched to the needs of spectroscopy, providing both convenient formats and high optical throughput. Conchello et al. and Svoboda et al. have summarized these techniques for optical imaging of thick tissues.^164,166 Some new optical sectioning techniques, such as planar illumination and structured illumination have been proposed in recent years. These methods are particularly good for applications involving high speeds, large FOVs, or long-term imaging, which have been reviewed by Mertz.¹⁶⁷ Structured illumination is a way to make it possible to perform optical sectioning with a conventional microscope, which is generally radiated through a grid of horizontal lines and is then reflected off a beam splitting mirror. Then the image of the thick sample with both wide-field and optical section containing grid lines passes back through the beam splitter to the camera. Thus, the optically sectioned images can be reconstructed by some algorithms from a conventional wide-field microscope.¹⁶⁸ Most of these techniques can also be used for spectral imaging of thick tissues. For example, Oruanaidh et al. proposed a Bayesian spectral analysis method to obtain optical sectioning using structured light imaging in a conventional wide-field microscope.¹⁶⁹ Webb et al. presented a wide-field fluorescence lifetime imaging system with spectral resolution and optical sectioning to achieve five-dimensional fluorescence images of biological tissues.¹⁷⁰ In addition, the digital micromirror device also has been employed as a spatial excitation light modulator for either spectral imaging or fluorescence lifetime imaging.^171,172 In recent years, single-point optical sectioning systems for microspectroscopic analysis have been reported, some of which have been adapted for scanning, particularly for Raman spectroscopy.¹⁷³ The optical sectioning spectroscopic systems using slit scanning have been applied both for transmission and Raman spectroscopic measurements also.¹⁷⁴ The programmable array microscope based on programmable spatial light modulators, developed by Hanley et al., is another kind of optical sectioning microscope of particular interest for spectroscopy.¹⁶⁵ These studies show that optical sectioning has improved spatial resolution in the intensity images by removing out-of-focus blur. With optical sectioning, a high spatial resolution image generally demands a greater signal-to-noise ratio (SNR) than a conventional intensity image does. This will make high S/N ratios imperative when applying the technique to thick biological tissues.¹⁷⁵ Therefore, suitable specification of the optical system is generally needed to produce optimal trade-offs between reducing the speckle noise and preserving the optical sectioning to get clear images. Some of these studies also demonstrate that the spectral imaging technology combined with optical sectioning is essential for spectroscopic characterization. For example, the spectral signature changes of tissues are generally associated with their pathologic states. The ability to localize these changes within cells and tissues will improve the diagnostic capabilities of such method. In some cases, the localization of probe molecules within a specimen requires that they can be distinguished from sample autofluorescence or from out-of-focus probe-fluorescence. Therefore, it is important for the spectral imaging microscope to reduce both types of background through the use of optical sectioning methods to improve the detection and accurate localization of targets.¹⁶⁵ However, there are no hard and fast prescriptions on which technique is most suitable for optical sectioning. The researchers must consider what factor is important in the biomedical analysis (such as speed, resolution, and FOV) as the optimization of one factor often leads to degradation of the others.¹⁶⁷

4.4.

System Calibration and Data Preprocessing

After the spectral imaging system was developed, one of the important steps was to ensure that the system would provide an accurate measurement of the spectrum. That is, the spectral imaging systems need to be calibrated and corrected to produce a flat spectral response before further experimental measurements. These steps are important for ensuring reproducible results and for comparing spectra across multiple systems.¹⁴¹ One way to determine the wavelength accuracy of the spectral imaging system is to calibrate it with a low-pressure ${\mathrm{Hg}}^{+}{/\mathrm{Ar}}^{+}$ discharge lamp that covers the wavelength range from 365 to 850 nm. However, the pure ${\mathrm{Hg}}^{+}{/\mathrm{Ar}}^{+}$ lamp usually emits deep UV light that is dangerous to the exposed skin and can cause blindness to unprotected eyes, thus making the measurement difficult to perform. Therefore, an eye-safety, multiion discharge wavelength calibration lamp (MIDL) distributed by LightForm Inc. can be used for wavelength calibration from 365 to 912 nm simultaneously.¹⁶³ Lerner and Zucker have proposed the MIDL calibration method in detail and have derived reference spectra from the MIDL data to accurately predict the spectral resolution, ratio of wavelength to wavelength, contrast, and the aliasing parameters of any confocal spectral imaging system.¹⁸⁴ This method has also been used by Leavesley et al. to calibrate the HSi-300 HSI system (ChromoDynamics, Orlando, FL).¹⁴¹ In addition, some other spectrally well-known light sources, such as lasers (for example, 632.8 nm by HeNe lasers), fluorescent lamps, and broadband lamps equipped with interference bandpass filters, can also be used for wavelength calibration.¹⁷ Other calibration methods also have been presented recently. For example, three National Institute of Standards and Technology traceable Spectralon standards (Labsphere, North Sutton, NH) of different colors have been used to calibrate the multiaperture system, which can capture six identical images of the human fundus at six different spectral bands.¹²² Xu et al. developed a digital tissue phantom (DTP) platform for quantitative calibration and performance evaluation of spectral wound imaging devices and demonstrated the feasibility of such DTP platform by both in vitro and in vivo experiments.¹³⁶ After the spectral calibration process, the wavelengths for the pixels along the spectral dimension of the data cube can be defined and the relationship between the true spectral position of the incoming light and the observed effect can be determined for further processing.

Besides the spectral calibration, a data preprocessing procedure is also needed to remove the effects of noises and artifacts and make the images suitable for quantitative analysis. During the spectral image acquisition, the noise counts accumulated on the detector may influence the pixel values beyond the true intensities. Various image artifacts can be generated by factors such as nonuniform illumination, dust on the lens surface, and pixel-to-pixel sensitivity variations of the detector.¹⁷ These effects in the original data can be corrected by some flat-field correction methods. In the past decades, different data correction methods have been proposed along with the development of the biomedical spectral imaging technology. Most of these methods utilize the radiance of standard white diffuse reference panels (for example, WS-1, Ocean Optics, FL) and the dark current data for the correction, which can be performed using the following equation:^{83,122,141,176,185}

Some commercial software has also been developed for biomedical spectral images analysis. For example, the CytoSpec program developed by Peter Lasch is a specialized software package with a large number of different functions, such as data import, spectral and spatial preprocessing, and uni- and multivariate image segmentation.¹⁸⁶ Some other software associated with the commercial spectral imaging hardware also provide the basic calibration and preprocessing functions, such as the inForm™ intelligent image analysis software developed for the Nuance™ and Vectra™ multispectral imaging system, the HyperSpect™ cross-platform operating software for the 710VP Hyperspectral Imager, and the PARISS® software for the PARISS® system.

5.1.

In Vivo Spectral Imaging

In vivo optical imaging allows a noninvasive observation of living organisms and helps in understanding metabolic processes and disease-related changes in the body. As one of the optical diagnostic technologies, spectral imaging has been used for in situ tissues diagnosis in recent years. Among the openly accessible tissues, skin is one of the most suitable ones for in vivo spectral imaging. Afromowitz et al. developed a multispectral imaging system to evaluate burn depth of skin.¹¹² They found that the spectral imaging method was more accurate in predicting burn healing than the traditional method. After that a spectral image-analysis system based on Fourier transformed spectroscopy combined with image processing has been used for the in vivo study of porphyrin localization in human skin lesions.¹⁰⁷ These are some earliest studies on in vivo analysis on skin. Then Sowa et al. evaluated the hemodynamic changes in the early postburn period of skin by NIR reflectance spectroscopy and imaging.^59,60,64 Cancio et al. further presented a study in which a spectral imaging system was used to assess the cutaneous manifestations of significant systemic events.⁶⁷ Martinez proposed a noninvasive spectral reflectance method for mapping blood oxygen saturation in various types of wounds, ranging from burn injuries and diabetic-related ulcers, to skin grafts.¹⁸⁷ Some other studies include determining skin ulcers,¹⁸⁸ and spectral imaging of cancer skin fluorescence and reflectance.^5,16,22 In vivo confocal scanning laser microscopy has also been used to analyze the skin annexes, as well as cutaneous cells from different epidermal layers by both reflectance and fluorescence mode.¹⁸⁹ More recent studies focused on 3-D HSI on small human skin areas,⁶⁸ quantification of skin oxygen saturation and its relation to the traditional measures of wound healing,¹²³ subcutaneous adipose tissue observation,¹⁷⁶ ischemic wound assessment,¹³⁶ and breast tumors.¹⁴² These studies show that noninvasive in vivo detection of skin surface using spectral imaging technology may be of value in the early assessment of burn injury and skin tumors.

Another important in vivo application of spectral imaging is retina detection. Some earlier studies focused on the development of the retinal spectral imaging system coupled with a standard ophthalmic fundus camera.^144,190 Some researchers also discussed the analysis methods in detecting and characterizing the extent of retinal abnormalities according to the spectral images.^41,191,192 A major disadvantage of the spectral imaging systems used in these studies is that they are fundamentally unsuitable for recording phenomena that are changing on a timescale that is shorter than the duration required in reconstructing the spectral data cube. Therefore, the snapshot spectral imaging technology has been proposed and used to capture retinal spectral images in recent studies. Harvey et al. developed an image replicating imaging spectrometer to capture spectral images of the retina and proposed a physical model that enables the oximetry map of retina.^114,116 Johnson et al. and Ramella-Roman et al. obtained oxygen saturation maps of retina with a three-wavelength algorithm and found that the oxygen saturation were $\sim 95\%$ for healthy subjects’ arteries and 30 to 35% less for the veins.^18,122 Although the snapshot mode can capture the retinal spectral imaging quickly, the spatial and spectral resolutions are limited. In recent studies, LCTF-based retinal HSI systems have been developed to calculate the oxygen saturation of retinal vessel and detect retinal lesions.^117,118,193 These studies show that a spectral imaging technique can be adapted to measure retina in vivo and map relative oxygen saturation in retinal structures. The confocal scanning laser ophthalmoscopy has also been used to evaluate retinal perfusion in the human eye.^137,138 In addition, it can be combined with high-resolution spectral-domain optical coherence tomography (OCT) for simultaneous multimodal imaging of retinal pathologies, such as fluorescein and indocyanine green angiographies, IR and blue reflectance (red-free) images, fundus autofluorescence images, and OCT scans.¹⁹⁴ Future work in this area will include clinical studies focused on spectrally characterizing retinal tissue, its diseases, and on the detection and tracking of the progression of retinal disease.^118,191

The in vivo spectral imaging technology has also been used for tongue surface identification and oral cancer diagnosis. Li et al. introduced the spectral imaging technology to tongue diseases diagnosis for the first time.³⁶ Yu et al. presented in vivo spectral images from a hyperkeratotic lesion on the ventral surface of the tongue to demonstrate clinical applicability.⁵⁴ In recent studies, different kinds of HSI systems have been developed and used for tongue body segmentation, tongue color analysis and discrimination, tongue cracks extraction and classification, sublingual veins analysis, and malignant changes associated with oral cancer detection.^{4,37–39,80,195} These preliminary experiments show that the spectral oral imaging system is superior to the traditional CCD-based methods because it can provide more information about the oral surface. However, further studies are needed on how to extract the quantitative features of the tongue surface and model the relationship between these features and certain diseases.

Besides the exterior organs and tissues, some intracorporeal organs also can be detected in vivo by the spectral imaging system coupled with colposcope, endoscope, and gastroscope. Parker et al. developed a noncontact in vivo hyperspectral diagnostic imaging device for the detection and localization of cervical intraepithelial neoplasia.¹³⁵ Balas described an LCTF-based multispectral imaging system for the in vivo early detection, quantitative staging, and mapping of cervical cancer and precancer.⁷⁷ Commercial systems such as the dynamic spectral imaging colposcope (DySIS, Dynamic Spectral Imaging System; Forth Photonics, Livingston, UK) have also been developed and used to detect cervical lesions.^157,158,196 Similarly, Vo-Dinh et al. developed an AOTF-based endoscopic HSI system and evaluated it by a model of bronchial tubes (APM Inc.) and esophageal cancer tissues.^{5,16,22,84,85} Arnold et al. evaluated the hyperspectral video endoscopy demonstrator during several measurements in a clinical surrounding for thorax surgeries.⁸⁷ In recent study, this technique has also been used in gastrointestinal diagnosis.⁸⁸ Although most of these studies were mainly concentrated on the system design, and only preliminary results have been obtained, they have demonstrated the possibility of inspecting inner regions of the human body with additional diagnostic information by spectral imaging technology. There are also some other in vivo applications of spectral imaging systems, such as hemoglobin saturation evaluation in tumor microvasculature and tumor hypoxia development,⁹¹ characterization of activity-dependent changes in oxyhemoglobim, deoxyhemoglobim, and light scattering in brain,¹⁹⁷ analysis of the oxygenation and blood flow through microvessel networks,¹⁹⁸ normal and malignant muscle tissues detection,⁸⁴ tooth decay detection,^82,113,185 and flow cytometry.⁸⁶ Guidance of surgical intervention, treatment, and real-time assessment of tissue response to therapy is another promising application of in vivo spectral imaging.^16,199 Especially with the development of clinically relevant quantum dots (QDs) contrast agents suitable for spectral multiplexing, in vivo spectral imaging can help the surgeon to delineate tumor margins and to detect residual tumor cells or micrometastases. Singhal et al. summarized the advances in nanotechnology and optical instrumentation and how these advances can be integrated for applications in surgical oncology applications, such as tumor localization, tumor margins, important adjacent structures identification, mapping of sentinel lymph nodes, and detection of residual tumor cells.²⁰⁰ From the review, it can be found that the combination of nanoparticles and in vivo spectral imaging has particularly promising opportunities in surgical oncology, and it will be an important research trend in the future studies.

In addition, some researchers also use the diffuse spectral imaging system to perform the in vivo diagnosis. This technology has been used for early diagnosis and screening of malignant changes in the oral cavity. Some experimental results show the potential use of this method in routine clinical practice for instantaneous screening of oral cancer.^201,202 Prasanth et al. investigated the in vivo inflammation mapping of periodontal disease using the diffuse reflectance spectral imaging method and estimated the specificity and sensitivity of the in vivo technique in inflammation mapping.²⁰³ Recently, some new kinds of diffuse reflectance spectral imaging methods have been proposed and used for normal and malignant breast tissue–mimicking phantoms.^204–206 All these studies show that the spectral imaging technology has the potential for in vivo diagnosis on diseases without the need for sample excision and histological processing. Both the spatial and spectral diagnostic information can be available noninvasively and in real time for medical diagnostic applications where rapid in vivo detection of complex samples is required. Therefore, this approach is expected to enhance the dynamic range of the image contrast in identifying the earlier changes of live tissues.

For the in vivo applications, the proper selection of wavelength range is important to match the identification of the object. For example, the spectral imaging method has been used to map the blood oxygen saturation in various types of surface wounds (such as burn injuries, diabetic-related ulcers, and skin grafts), retinal structures, and the blood vessels of a living animal as we have mentioned above. The main basis of this application is that the hemoglobin is found primarily in two forms in the blood, that is, oxyhemoglobin (${\mathrm{HbO}}_2$) and deoxy-hemoglobin (Hb). Light that emerges from the body is modified by the absorption of hemoglobin and contains information about the amount of oxygen contained in blood. There are usually different spectral signatures from 500 to 800 nm between the spectra of ${\mathrm{HbO}}_2$ and Hb, which can be used for oxygen saturation detection.²⁰⁷ One of the wavelength regions between 700 and 800 nm can be used for mapping the blood oxygen saturation in skin surface as the states of hemoglobin at an isosbestic (800 nm) and oxygen-sensitive (760 nm) wavelength are different. In addition, it allows for deeper penetration of light at the longer wavelengths and minimizes the influence of other absorption chromophores below 630 nm.¹⁸⁷ For the blood oximetry in the retina, wavelengths ranging between 500 and 650 nm can be selected. The reason is that the optical depth of ${\mathrm{HbO}}_2$ changes from being less than to being greater than the thickness of the larger retinal arteries from 500 to 650 nm, and the reflectance spectrum of different locations across the retina is dominated by ${\mathrm{HbO}}_2$ and Hb in both choroidal and retinal vessels, and by melanin components that vary depending on subjects and ethnicity.^114,117 Another example is tooth diseases’ diagnosis using the spectral imaging method in the NIR spectral range. Studies show that dental enamel manifests high transparency (the mean free path of the photons being $\sim 3\mathrm{mm}$ at 1310 and 1550 nm, while being $<0.1\mathrm{mm}$ for visible wavelengths) for the light in the NIR spectral region and the tooth caries process usually increases absorption and scattering of the incident light. Therefore, the NIR spectral range (for example, 900 to 1700 nm) is suitable for noninvasive high-contrast imaging of the caries located both beneath the enamel surface and on the surface between adjacent teeth.^82,113 For the clinical study of the uterine cervix, spectral imaging and analysis show that the maximum contrast between acetic acid responsive and nonresponsive areas can be obtained at $525\pm 25\mathrm{nm}$. Therefore, this spectral range can be selected to get the maximum diagnostic information for mapping the cancerous and precancerous lesions of cervix.²⁰⁸ These studies demonstrate that it is important to select the proper wavelength range for the spectral imaging system in the in vivo biomedical applications as different biological changes usually have different spectral signatures at different wavelengths.

Since spectral imaging technology can collect a set of images at continuous wavelengths revealing the spectral characteristics of multiple tissue components, the in vivo applications of this technology opens a new avenue for noninvasive and real-time assessment of the tissues. One advantage of the in vivo spectral imaging is that both the functional and the structural pathologic changes of the tissues can be detected and used for the early diagnosis and grading of diseases without the need of a sample excision. Another advantage is that the spectral imaging systems used for in vivo detection can support a wider wavelength range (including the NIR band) than those designed for histological applications, as the front optics are usually simple and easy to change. However, spectral imaging has a significant limitation when used for in vivo detection, that is, the penetration depth, due to the strong scattering of optical radiation in tissues. Therefore, those openly accessible sites (such as skin, retina, cervix, tongue, teeth, and tumor microvasculature) and endoscopy-accessible organisms (such as stomach and colon) are suitable for in vivo spectral imaging.

5.2.

Fluorescence and Cytogenetics

In traditional fluorescence detection methods, it is necessary to excite all the dyes, but the emission spectral range cannot overlap with the excitation range. This means that part of the spectral range that is used for excitation cannot be used for detection. With the spectral information, spectral imaging technology makes it possible to resolve the spectra of fluorescence images without switching optical filters and unmix the cross-talk between overlapping fluoresence emissions.²⁰⁹ One of the most widespread applications of spectral imaging technology is SKY, which is based on FISH and revolutionized mouse and human cytogenetics. Schrock et al. developed a spectral imaging system based on Fourier spectroscopy and used it to resolve spectrally overlapping chromosome-specific DNA probes and identify all human chromosomes simultaneously.¹¹⁰ This is the earliest study on multicolor SKY of human chromosomes with $>1100$ citations since first published in 1996. Full karyotype single interphase cells have also been implemented by a rapid enumeration procedure based on the hybridization of chromosome-specific probes and spectral imaging analyses.²¹⁰ SKY was recently used for evaluation of chromosomal rearrangements in post-Chernobyl papillary thyroid carcinomas,²¹¹ identification of constitutional chromosomal abnormalities,²¹² and characterization of giant marker and ring chromosomes in a pleomorphic leiomyosarcoma.²¹³ These studies show that spectral imaging technology has the ability to identify and detect complex karyotypes, subtle chromosome translocations, and aberrations of chromosomes that have been extracted from a metaphase nucleus and hybridized with chromosomes probes, without spectra overlapping. Therefore, it is a valuable tool for the cytogenetic analysis of a broad range of chromosome abnormalities associated with a large number of genetic diseases and malignancies.²¹⁴ There are also a number of different labeling strategies that have been combined with SKY, such as the fluorescent dyes [for example, Rhodamine, Texas Red, Cy5, fluorescein isothiocyanate (FITC), and Cy5.5], the fluorescent proteins, QDs, and some commercial products (such as the fluorescent polymer “Brilliant Violet” series, a set of FISH-probes developed by the Applied Spectral Imaging, ASI). With the appropriate spectral image capturing and analysis methods and tools (for instance, SpectraCube™ and SKYView™, ASI), these probes can represent all 24 chromosomes (SKY Paint) together.^20,215

The spectral imaging technology also provides a unique capability for detecting spectral changes in fluorescent indicators [for example, an indicator for fluorescence resonance energy transfer (FRET)], microenvironmental changes in cellular compartments, fluorescence from natural pigments, and effective detection of multiple fluorescent labeling in biological tissues. For example, Haraguchi et al. studied the capability of the spectral imaging system for the spectral and temporal resolution of fluorescently stained living cells and found that the linear unmixing of the spectral images can separate the cyan fluorescent protein (CFP) staining of the nuclear membrane and yellow fluorescent protein (YFP) staining of STAT1 in some cases.²¹⁶ However, the traditional FRET sensors were limited in the range of glutamine concentration that can be detected due to relatively low affinity, and the commonly used FRET pair [enhanced CFP and YFP (ECFP and EYFP)] is difficult to multiplex with other FRET pairs. Some new fluorophores, such as the teal fluorescent protein mTFP1, may overcome these limitations as it provides a narrow and single-peaked emission spectrum, improved brightness and photostability, and single exponential lifetime decay. The new FRET pair, mTFP1 and venus, has also been proposed and used to analyze the properties of transporters expressed in a given cell as it can improve quantum efficiency and pH stability compared to the more commonly used FRET donor, ECFP and EYFP.²¹⁷ The mTFP1–venus FRET pair is suitable as a replacement for ECFP and EYFP and can be duplexed with another FRET pair, mAmetrine and tdTomato, opening up the possibility for real-time imaging of another molecule.²¹⁸ The time course of fluorescence from a coral during a photobleaching experiment has been observed by using the spectral imaging method. It provides a hint of the photophysics of the intact coral leading to a better understanding of the likely individual functions of the many fluorescent and nonfluorescent proteins following a green fluorescent protein (GFP)-like motif isolated from different varieties of coral.²¹⁹ Studies by Livet et al. show that $>90$ different neuronal clones could be distinguished in the inventors of Brainbow mice according to color and intensity gathered by combining tissue/temporally inducible Cre recombinase, multiplex recombination sites encompassing multiple fluorescent protein gene cassettes, and spectral confocal microscopy.²²⁰ In addition, the spectral confocal microscopes have also been used for fluorescence evaluation and quantification in different tissues, such as brain slices, mouse ovaries, skin sections, kidney slice, and breast cancer tissues.^134,221,222 Neher et al. demonstrated that quantitative FRET can be performed by taking advantage of steady-state intensity measurements, but it only allows determination of the apparent FRET efficiency, the relative concentration of donor, and the concentration of the acceptor.²²³ They also developed a nonnegative matrix factorization (NMF) algorithm to detect and separate spectrally distinct components of multiple labeled fluorescence images.¹⁸² In addition, Fisher et al. examined the fluorescence changes in the human oral tissues by a Sagnac Fourier transform imaging spectroscopy combined with a fluorescence microscopy.²²⁴ In cell signaling, Grant et al. presented an approach for simultaneous imaging of two different FRET sensors in the same cell with minimal spectral cross-talk, which can be used to report on both unimolecular FRET biosensors and protein–protein interactions in the same cell.²²⁵ Studies by Kawano et al. show that a combination of the coiled-coil tag-probe labeling method and spectral imaging can enable a stoichiometric analysis of the oligomeric state of membrane proteins on living cells by using monomeric, dimeric, and tetrameric standard membrane proteins.²²⁶ Draux et al. tested different optical substrates that would best preserve cell integrity and allow direct measurement of Raman spectra at the single living cell level. They also performed Raman spectral imaging to show the distribution of nucleic acids, lipids, and proteins within the cell.²²⁷ There are also some other applications that have been demonstrated in separating multiple fluorescence signals,^120,141 separating autofluorescence and fluorescence signals,^12,85 extracting dye distributions from the images,²²⁸ FRET imaging,^53,229 and tissue imaging.⁴⁵ These studies show that spectral imaging is best appreciated as a tool for detecting spectral changes in fluorescence and eliminating problems associated with autofluorescence. It extends the possibilities in discriminating distinct fluorophores with highly overlapping emission spectra, thus providing the possibilities of multicolor imaging.²²⁹

There are also some requirements for the spectral imaging system in detecting and separating fluorescence signals. Theoretical analyses on these requirements have been done along with the development of spectral imaging fluorescence microscopy. Garini et al. presented the SNR to evaluate the multiple-color fluorescence microscopy method for the particular case of SKY.²³⁰ R. Neher and E. Neher discussed the influence of photobleaching on the resolution and on the choice of imaging parameters for optimal fluorophore separation. They introduced a new measure called FoM for the efficiency of a given microscope or measurement configuration and found that a small number of spectral channels (two to four) is sufficient to extract spectral information and works almost as well as when using many channels, as long as the spectral ranges are optimally selected.¹⁸¹ They also provided theoretical analysis on how to separate fluorescence of FRET pairs from those originating from unpaired donors and acceptors by spectral fingerprinting and on how to select imaging parameters to optimize the SNR of the estimates.²²³ These theoretical analyses are very important for evaluating the fluorescence separation ability of spectral imaging fluorescence microscopy. With these theoretical analyses, R. Neher and E. Neher found that for the autofluorescent proteins ECFP, enhanced green fluorescent protein, and EYFP and two sequential exposures at 458 and 514 nm, the FoM is $\sim 0.4$, which means that $\sim 2.5$ times more photons should be excited and detected to reach the resolution than if the dyes were not mixed and if their fluorescence could be measured without background noise.¹⁸¹ Quantitative analysis of fluorescently labeled cells in highly autofluorescent lung vasculature revealed that GFP may be detected when the GFP signal contribution is as little as 15% of the peak autofluorescence intensity.¹⁴¹ For a spectral imaging system with 7.2 nm wavelength resolution, at least a 90% confidence level can be obtained to resolve all fluorochrome dyes that have an FWHM $>40\mathrm{nm}$ and whose maxima are at least $\sim 7\mathrm{nm}$ apart, such as Cascade Blue, 4’, 6-diamidino-2-phenylindole (DAPI), GFP, FITC, Rodamine 123, GFP, Cy3, tetramethyl rhodamine isothiocyanate (TRITC), Lissamine Rodamine, Texas Red, and Cy5.⁴² The number of the labels that can be used simultaneously generally depends on the excitation and emission paths, the spectral properties and stability of the fluorophores, quantum efficiency and the noise properties of the system and the amount of confounding autofluorescence, etc. In general, the degree of multiplexing is 4 to 5 in immunohistochemistry and $\sim 10$ in immunofluorescence.^44,231 The spectral range selection of a spectral imaging system is also important in fluorophores separation, as it is essential in distinguishing the weak emitted light from the strong excitation light at different wavelengths. The fluorochromes have to be excited at certain spectral ranges, which limit the spectral ranges available for detection. For instance, better spectral unmixing of gastric tissues can be got by using excitation rather than using emission spectra. This is because wavelength-dependent variations in tissue absorption and scattering over the near infrared range are far less than in the visible range.²³² Overall, a spectral range from 400 to 750 nm usually has been selected for fluorescent studies as the excitation spectrum of most of the fluorescent probes are located in this wavelength region. In addition, as time-dependent photobleaching will alter the measured spectral signature, there are also limits to the maximum acquisition time tolerated by the spectral imaging fluorescence microscopy. Wachman et al. has discussed the imaging speed and spectral versatility of AOTF microscopy in the investigation of millisecond and submillisecond dynamics in living biological systems using multiparameter fluorescence. They found that polystyrene beads brightly labeled with Cy3, Cy5, and both Cy3 and Cy5 can be separated according to the image set acquired in 3.5 ms.⁹⁹ In general, there is a trade-off between achieving a high identification accurate through increased image acquisition time and minimizing spectral photobleaching artifacts. Therefore, the parameters should be properly selected according to the characteristics of different fluorescence probes.

These applications and analyses demonstrate that one of the advantages of spectral imaging compared to conventional microscopy methods is the increased number of distinct fluorescent molecules that can be simultaneously detected and distinguished. For example, by knowing the fluorescence spectra of each of the dyes, up to seven observed mixed spectra can be unmixed to each component by the spectral imaging technology.¹¹¹ Another advantage is its discriminatory abilities, as the linear unmixing analysis often results in improved SNR of the system. Compared to single-channel images, there is commonly a two- to threefold increase in SNR after unmixing.¹⁴¹ In addition, the advent of spectral imaging will also provide an impetus for the development of a new class of fluorescent dyes. There are, however, some limitations induced by factors, such as image background and detector noise, and the appropriate selection of the number and bandwidth of the detection channels with respect to the overlap of the fluorescence to be distinguished. These limitations become most critical when imaging living cells, where the signals of interest are usually weak. So far it is parallel and thus extends not only to the abilities and performance of multicolor imaging, but also to the use of more efficient donor–acceptor fluorophore pairs for FRET imaging.¹⁶⁰ Moreover, an inherent limitation is that quantification improvements rely heavily not only on prior knowledge of the absorption spectra of the tissue chromophores and the magnitude of their relative contributions. In cases where this information is not available, modeling errors can propagate into the inversion process, potentially negating any gains attributable to the spectral constraints.²³³ Another limitation of the fluorescence spectral microscopy is that the observed fluorochromes have to be excited at certain spectral ranges, limiting the spectral range available for detection. In addition, both the emission and the absorption spectra of simultaneously present dyes overlap and the signal quality is limited by photobleaching. This problem has been formulated by Zimmermann et al. in the case of live-cell spectral imaging, which introduces even stronger restrictions on the acquisition time.²²⁹ A more detailed method for solving this problem has been proposed by R. Neher and E. Neher also in the case of multiple fluorochromes by taking into account the bleaching effects.¹⁸¹ Despite this, spectral imaging microscopy and corresponding spectral analysis are useful methods for detecting weak fluorescent signals within a high autofluorescence background.

5.3.

Histological Analysis

Histology is an essential tool of biology and medicine, which is commonly performed by examining cells and tissues by sectioning and staining, and followed by structural examination under a light microscope. To visualize different components of the tissue under a microscope in bright field, the sections are usually dyed with one or more stains. The histological staining of tissue sections enables the microscopic visualization of tissue structures and substances onto which the stains are attached without the need for selective spectral excitation as in fluorescence. It has been widely used in pathology, where the visual assessment of complex staining patterns constitutes the basis for the identification and classification of the status of the examined tissue.⁷⁸ However, the visual assessment by the traditional light microscope can usually obtain qualitative and subjective results, which leads to only part of the diagnostic information of the stained tissues being explored.²³⁴

The spectral imaging technology may provide a new tool for histological analysis. It can take advantage of the spectral features of various stains used in histological sections to segment features of interest for subsequent morphometric measurement. In recent years, the method has been successfully applied to a wide range of histological stains and immunolabeling reaction products and has proved capable of detecting any color-stained object. Ornberg et al. presented spectral analysis of stain combinations such as Safranin O/hematoxylin for proteoglycans in articular cartilage, Oil red O in adipose tissue, peroxidase-diaminobenzi-dine immunostain with either fast green or hematoxylin counterstain, alkaline phosphatase-AP-Red, and -Vector Red immunostain and counterstains in a variety of tissues.⁶⁵ In addition, Papadakis et al. discussed the possibility of using spectral imaging method for quantitative assessment of the uptake by histologic samples of stains used in pathology to label tissue features of diagnostic importance.⁷⁸ This method has also been used to identify emissions on printed DNA microarrays and tissue microarrays.^26,235 Li et al. presented the analysis method on spectral images of retina sections dyed with hematoxylin for evaluating the protective effect of erythropoietin on diabetic retinal cells and the spatial-spectral method for automated white blood cells segmentation.^49,97 Kiyotoki et al. developed an objective method to detect gastric cancer using HSI technology combining spectroscopy and imaging.²³⁶ Larsen et al. presented the hyperspectral image analysis combined with the histology method for detection and characterization of advanced atherosclerotic plaques in vitro.²³⁷ Maeder et al. introduced a confocal laser-scanning microscope equipped with a detection unit that provides high spectral resolution method to visualize and evaluate the spatial distribution of spectral information in skin sections.²²¹ In a more recent study by Nallala et al., an IR spectral imaging system was used for histopathological recognition in colon cancer diagnosis.²³⁸ Fourier transform IR spectral imaging system has also been used to perform the histopathology mapping of biochemical changes in myocardial infarction, live epithelial cancer cells, histologic and biochemical correlations in tissue engineered cartilage, and histomorphological evaluation of oral squamous cell carcinoma of the oral cavity.^109,239 Moreover, the spectral imaging technology allows the investigator to separate dyes with overlapping emission spectra and to eliminate from the display certain dyes in a sample.¹³ In some cases, spectral images of unstained tissue sections may also provide useful information similar to the results available from stained histological methods.^98,240,241 Potentially, this is a way to identify abnormalities within tissue at an earlier stage disease, as a change to the molecular composition of a cell most likely occurs before a morphological change. Furthermore, some Raman spectral imaging methods also have been introduced by researchers for different kinds of biological analysis, as it can visualize molecular vibration with high sensitivity and high contrast, allowing label-free imaging of biological samples. For example, Amharref et al. investigated molecular changes associated with glioma tissues by Raman microspectroscopy and then compared the spectroscopic markers obtained from C6 glioma tissues with conventional histological and histochemical techniques.²⁴² Tollefson et al. evaluated whether Raman molecular imaging could be used to identify histological differences in patients with Gleason 7 prostate cancer who progressed to metastatic disease and died from prostate cancer.²⁴³ More recent study by Ozeki et al. has used the Raman scattering microscopy in various imaging modalities for histological analysis, such as 2-D spectral imaging of the rat liver, spectral imaging of several sections of intestinal villi in mice, and in vivo spectral imaging of mice ear skin.²⁴⁴ Kong et al. have developed electro-optical tunable Lyot filter–based multicolor stimulated Raman scattering microscopy for live HeLa cells analysis.²⁴⁵ In addition, there are also some commercial spectral imaging systems that have been used for visualization and quantitative analysis of multicolor tissue sections, for instance, the Nuance™ multispectral imaging system and morphologic image analysis packages, such as the inForm™ image analysis software (CRi), used for assessing morphological distributions of miRNAs and their putative targets in thin tissue sections,¹⁶⁰ the Prism and Reflector Imaging Spectroscopy System (PARISS®) hyperspectral imaging system (LightForm) used for studying apoptosis,²⁴⁶ and the HSi-300 hyperspectral imaging system (Gooch & Housego) used for preclinical research and clinical pathology.²³¹

These studies show that the histological spectral imaging method can improve both morphometric and biochemical analysis by enabling the evaluation of subtle cytological and histological features, which in turn can yield useful structural and medical information for diagnostic and prognostic evaluation. Moreover, the spectral imaging method can also identify the combination of stains by determining the spectral signatures of each stain on the slide and can select two or more wavelengths at which the differential absorption between stain and counterstain is greatest. It allows researchers to analyze multistains in bright-field mode and calculate the concentration of each and separate the complex color-image into a set of single color images, where each one of them shows the sample as if it were stained with only a single stain. One advantage of histological spectral imaging is that many stains can be used simultaneously, while still making it possible to analyze the histological section as if it were stained with a set of single stains. Another advantage is that it can provide information on the colocalization of structures, proteins, and other components by correlating the single stain distribution.¹⁶² To identify the stained objects in histological sections, the wavelength ranges should be selected to cover red (R), green (G), and blue (B) light, i.e., at wavelengths of $\sim 630$, 545, and 435 nm, respectively.⁶⁵ The reason is that most of the commonly used dyes are selected to highlight the stained objects of interest above a uniform background to match the spectral response of the human eye, which restricts in visible light. Therefore, a wavelength range covering the R, G, and B light, such as 400 to 650 nm, is usually needed for the histological applications. Another requirement for histological application is the proper spectral resolution. Spectral analysis is based on the ability of histological and cytological specimens to absorb, reflect, or emit photons in ways characteristic to their interactions with specific stains, whose characteristic absorption and emission bands usually occur in 10 nm or less narrow-wavelength ranges.²¹⁵ Therefore, the spectral resolution of the spectral imaging systems generally needs to be $<10\mathrm{nm}$ to enable the demarcation of features and the evaluation of quantities from the spectral information of the histological specimen.

In traditional histological analysis, a good visual contrast in double immunoenzyme staining plays an essential role for identification by unaided eyes or RGB camera. In some cases, an RGB camera is sufficient to distinguish the two basic colors of double staining results. However, typical stain combination colors such as red/blue and turquoise/red have the drawback of containing one diffusely localized and rather insensitive/inefficient chromogen, which makes it hard to perform successful double staining with antibodies that show obscure and weak staining by nature or staining of tiny cellular constituents.²⁴⁷ With the spectral imaging technology, one can unmix the original observation according to the different spectral characteristics of chromogens and nuclear counterstain rather than the visual contrast. Although the histological spectral imaging method is well founded on the physical principles of light absorption and is simple and straightforward to perform, there are still some difficulties to quantitatively unmix up to four different markers in a few cases. This may be due to the fact that spectral imaging does not work ideally with transparent chromogens, and the number of stains present in the sample as well as the characteristics of the stains themselves will affect the performance of spectral imaging analysis methods.^247,248 With the improvement of system performance and the advance of new identification algorithms, this problem may be solved. Moreover, it has been recognized that the quantitative assessment of the complicated histological staining patterns is sometimes ineffective and that its qualitative nature is capable of only partially exploiting the diagnostic content of the tissue staining techniques.⁷⁸ One of the reasons is that histological dyes are usually difficult to control in terms of intensity of stain, from cell to cell and from section to section. As most staining reactions involve a chemical union between dye and the stained substance, the degree of stained tissue sections could be determined by conditions such as the staining time, the temperature, or pH of the solution. Therefore, the intensity of pathological images acquired using different time or facilities varies greatly depending on circumstances.²⁴⁹ These factors will make the quantitative assessment of the concentration of single or of multiple stains hard or not straightforward. Different from those histological dyes, some immunohistochemical reagents have the potential to support quantitative analysis. For example, the alkaline phosphatase (AP) and horseradish peroxidase are enzymes that are allowed to run as endpoint assays, which establish the level of a quantifiable effect indicative of a biologic process. An immunohistochemical dye is usually analogous to an enzyme-linked immunosorbent assay (ELISA) test performed in the clinical lab, and ELISA tests are widely recognized as being truly quantitative.²⁵⁰ Although immunohistochemical analysis has some practical advantages over immunofluorescence, the optical properties of the chromogens will affect the linearity and dynamic range of the assay. For instance, the popularly used brown diaminobenzidine (DAB) stain exhibits scattering behavior similar to that of melanosomes and does not obey the Beer-Lambert law (BLL), which underlies that the unmixing approach applies only to pure absorbers. In fact, it is hard to separate DAB from melanin pigmentation spectrally since their spectra arise from the same optical properties.²⁵⁰ Other chromogens used mostly make precipitating products, such as Vector Red, and also have a similar problem as it may not obey the BLL. However, this does not mean that quantitative immunohistochemistry is out of the question if the entire process, from sample acquisition to analysis and interpretation, is properly integrated, standardized, and combined with spectral imaging technology. For example, the Nuance™ system and inForm™ software (CRi) have been used for automated quantitative scoring of immunohistochemistry on tissue microarray recently.²⁵¹ With the advances of spectral imaging in immunohistochemistry, it will become a routine assay for the assessment of therapeutic selection markers, which are likely to increase in number and be applicable to some diseases.